4 WUs on a GTX 480


Advanced search

Message boards : Cruncher's Corner : 4 WUs on a GTX 480

AuthorMessage
Profile [B^S] thierry@home
Avatar
Send message
Joined: Feb 19 05
Posts: 35
Credit: 1,250,712
RAC: 0
Message 110361 - Posted 10 Feb 2011 16:23:41 UTC

    Last modified: 10 Feb 2011 16:26:48 UTC

    Just as an info, I run now 4 WUs at a time on a GTX 480 :

    - GPU Load: +/- 84%
    - Memory used: +/- 1350 Mb
    - T°: 68°C
    - Fan speed : 78%
    - Run time: +/- 95 minutes

    Cool :-)
    ____________


    Intel I7 930 - GTX 480 - Windows 7 64

    Join BOINC Synergy, the best team in the galaxy!

    Profile astrocrab
    Send message
    Joined: Jan 28 08
    Posts: 185
    Credit: 65,716,770
    RAC: 1,435
    Message 110366 - Posted 10 Feb 2011 20:10:29 UTC - in response to Message 110361.

      and what is gpu load for 3, 2 and 1 WU at a time?

      Profile [B^S] thierry@home
      Avatar
      Send message
      Joined: Feb 19 05
      Posts: 35
      Credit: 1,250,712
      RAC: 0
      Message 110369 - Posted 10 Feb 2011 22:42:09 UTC - in response to Message 110366.

        Between 62% (1) and 84% (4).
        ____________


        Intel I7 930 - GTX 480 - Windows 7 64

        Join BOINC Synergy, the best team in the galaxy!

        Profile astrocrab
        Send message
        Joined: Jan 28 08
        Posts: 185
        Credit: 65,716,770
        RAC: 1,435
        Message 110370 - Posted 11 Feb 2011 0:44:01 UTC - in response to Message 110369.

          you have to check 2 and 3 too, because i have info what 2 and 3 WU make the same load for GPU as 4. if so, there is no reason to make more than 2 WU.

          Profile Gary Roberts
          Forum moderator
          Send message
          Joined: Feb 9 05
          Posts: 2928
          Credit: 821,259,689
          RAC: 1,605,756
          Message 110372 - Posted 11 Feb 2011 3:28:00 UTC - in response to Message 110361.

            Just as an info, I run now 4 WUs at a time on a GTX 480 :

            How many CPU cores are used in support? In other words, how many concurrent CPU-only tasks (of anything) are you running? Is the elapsed time for crunching a CPU task (the wall clock time - not reported CPU time) being significantly affected? Have you actually measured wall clock time to see how well it agrees with what is reported for Run time?

            What settings do you use for <avg_ncpus> and <max_ncpus>

            - Run time: +/- 95 minutes

            So, every ~95 minutes, you complete 4 CUDA tasks? Pretty impressive!

            My keen interest stems from the fact that I don't (yet) own any CUDA capable cards but I do want to crunch the Parkes work (efficiently - ie not on a CPU) for as long as it lasts. In terms of Australian distances, that facility is just down the road from where I live :-).

            I have 12 ATI 4850s crunching MW and I'm considering an investment in some suitable CUDA capable units in the interim while I wait to see if there is a successful port to OpenCL. The commentary about an OpenCL version has moved from "unlikely any time soon - if ever" (admittedly made quite a while ago) to "perhaps later in 2011" made about 9hrs ago. The reality will probably be "some time in 2012" :-). With your type of numbers being bandied about, it's very tempting to put more pressure on an already overblown budget :-).

            ____________
            Cheers,
            Gary.

            Profile [B^S] thierry@home
            Avatar
            Send message
            Joined: Feb 19 05
            Posts: 35
            Credit: 1,250,712
            RAC: 0
            Message 110377 - Posted 11 Feb 2011 7:49:53 UTC - in response to Message 110372.

              Last modified: 11 Feb 2011 8:05:05 UTC

              I have an I7 CPU. It runs (for the moment):

              - 4 * Einstein (0,20 CPUs + 0,25 NVIDIA GPUs) +
              - 4 * Spinhenge WUs at 80% +
              - 4 * Spinhenge at 100%

              But the 0,20 CPUs, following the Tasks Manager, is in fact 0,40 CPUs.

              Yes, 4 WUs every 95 minutes

              The avg_ncpus is 0.2. But as I'm not a specialist, I use app_info file copied somewhere. I only changed the CUDA count.
              Here's my app_info file:

              <app_info>
              <app>
              <name>einstein_S5GC1HF</name>
              <user_friendly_name>Global Correlations S5 HF search #1</user_friendly_name>
              </app>
              <app>
              <name>einsteinbinary_BRP3</name>
              <user_friendly_name>Binary Radio Pulsar Search</user_friendly_name>
              </app>
              <file_info>
              <name>einstein_S5GC1HF_3.06_windows_intelx86__S5GCESSE2.exe</name>
              <main_program/>
              </file_info>
              <file_info>
              <name>einstein_S5R6_3.01_graphics_windows_intelx86.exe</name>
              <executable/>
              </file_info>
              <file_info>
              <name>einsteinbinary_BRP3_1.05_windows_intelx86__BRP3cuda32.exe</name>
              <executable/>
              </file_info>
              <file_info>
              <name>einsteinbinary_BRP3_1.00_graphics_windows_intelx86.exe</name>
              <executable/>
              </file_info>
              <file_info>
              <name>cudart_xp32_32_16.dll</name>
              <executable/>
              </file_info>
              <file_info>
              <name>cufft_xp32_32_16.dll</name>
              <executable/>
              </file_info>
              <file_info>
              <name>db.dev.win.96b133b1</name>
              </file_info>
              <file_info>
              <name>dbhs.dev.win.96b133b1</name>
              </file_info>
              <app_version>
              <app_name>einsteinbinary_BRP3</app_name>
              <version_num>105</version_num>
              <platform>windows_intelx86</platform>
              <avg_ncpus>0.200000</avg_ncpus>
              <max_ncpus>1.000000</max_ncpus>
              <plan_class>BRP3cuda32</plan_class>
              <api_version>6.13.0</api_version>
              <file_ref>
              <file_name>einsteinbinary_BRP3_1.05_windows_intelx86__BRP3cuda32.exe</file_name>
              <main_program/>
              </file_ref>
              <file_ref>
              <file_name>cudart_xp32_32_16.dll</file_name>
              <open_name>cudart32_32_16.dll</open_name>
              <copy_file/>
              </file_ref>
              <file_ref>
              <file_name>cufft_xp32_32_16.dll</file_name>
              <open_name>cufft32_32_16.dll</open_name>
              <copy_file/>
              </file_ref>
              <file_ref>
              <file_name>einsteinbinary_BRP3_1.00_graphics_windows_intelx86.exe</file_name>
              <open_name>graphics_app</open_name>
              </file_ref>
              <file_ref>
              <file_name>db.dev.win.96b133b1</file_name>
              <open_name>db.dev</open_name>
              <copy_file/>
              </file_ref>
              <file_ref>
              <file_name>dbhs.dev.win.96b133b1</file_name>
              <open_name>dbhs.dev</open_name>
              <copy_file/>
              </file_ref>
              <coproc>
              <type>CUDA</type>
              <count>0.25</count>
              </coproc>
              <gpu_ram>220200960.000000</gpu_ram>
              </app_version>
              <app_version>
              <app_name>einstein_S5GC1HF</app_name>
              <version_num>306</version_num>
              <platform>windows_intelx86</platform>
              <plan_class>S5GCESSE2</plan_class>
              <api_version>6.13.0</api_version>
              <file_ref>
              <file_name>einstein_S5GC1HF_3.06_windows_intelx86__S5GCESSE2.exe</file_name>
              <main_program/>
              </file_ref>
              <file_ref>
              <file_name>einstein_S5R6_3.01_graphics_windows_intelx86.exe</file_name>
              <open_name>graphics_app</open_name>
              </file_ref>
              </app_version>
              </app_info>
              ____________


              Intel I7 930 - GTX 480 - Windows 7 64

              Join BOINC Synergy, the best team in the galaxy!

              Profile [B^S] thierry@home
              Avatar
              Send message
              Joined: Feb 19 05
              Posts: 35
              Credit: 1,250,712
              RAC: 0
              Message 110378 - Posted 11 Feb 2011 8:20:22 UTC - in response to Message 110370.

                In fact the GPU load is:

                - 41% for 1 WU
                - 64% for 2 WUs
                - 72% for 3 WUs
                - 82% for 4 WUs
                ____________


                Intel I7 930 - GTX 480 - Windows 7 64

                Join BOINC Synergy, the best team in the galaxy!

                Profile astrocrab
                Send message
                Joined: Jan 28 08
                Posts: 185
                Credit: 65,716,770
                RAC: 1,435
                Message 110379 - Posted 11 Feb 2011 8:26:25 UTC - in response to Message 110377.

                  Last modified: 11 Feb 2011 8:28:42 UTC

                  what about gpu load for 2 and 3 WUs at a time?
                  upd: ok. thank's. sad, what my gtx 560 can't hande more than 2 WU because 1024 is not enough even for 3

                  Profile [B^S] thierry@home
                  Avatar
                  Send message
                  Joined: Feb 19 05
                  Posts: 35
                  Credit: 1,250,712
                  RAC: 0
                  Message 110380 - Posted 11 Feb 2011 8:28:15 UTC - in response to Message 110379.

                    The GPU load is just above your post
                    ____________


                    Intel I7 930 - GTX 480 - Windows 7 64

                    Join BOINC Synergy, the best team in the galaxy!

                    Profile astrocrab
                    Send message
                    Joined: Jan 28 08
                    Posts: 185
                    Credit: 65,716,770
                    RAC: 1,435
                    Message 110381 - Posted 11 Feb 2011 8:29:01 UTC - in response to Message 110380.

                      ok. thank's. sad, what my gtx 560 can't hande more than 2 WU because 1024 is not enough even for 3

                      Profile [B^S] thierry@home
                      Avatar
                      Send message
                      Joined: Feb 19 05
                      Posts: 35
                      Credit: 1,250,712
                      RAC: 0
                      Message 110382 - Posted 11 Feb 2011 8:33:28 UTC - in response to Message 110381.

                        Last modified: 11 Feb 2011 8:34:11 UTC

                        The memory used is:

                        - 433 Mb for 1 WU
                        - 710 Mb for 2
                        - 1000 Mb for 3 (from 965 to 1035)
                        - 1300 Mb for 4
                        ____________


                        Intel I7 930 - GTX 480 - Windows 7 64

                        Join BOINC Synergy, the best team in the galaxy!

                        Profile astrocrab
                        Send message
                        Joined: Jan 28 08
                        Posts: 185
                        Credit: 65,716,770
                        RAC: 1,435
                        Message 110383 - Posted 11 Feb 2011 8:39:23 UTC - in response to Message 110382.

                          The memory used is:
                          - 1000 Mb for 3 (from 965 to 1035)


                          what's why i can't run 3 =(
                          now i think to sell my 1024mb version and buy 2048 mb

                          mikey
                          Avatar
                          Send message
                          Joined: Jan 22 05
                          Posts: 966
                          Credit: 21,069,812
                          RAC: 177,307
                          Message 110386 - Posted 11 Feb 2011 10:47:25 UTC - in response to Message 110383.

                            The memory used is:
                            - 1000 Mb for 3 (from 965 to 1035)


                            what's why i can't run 3 =(
                            now i think to sell my 1024mb version and buy 2048 mb


                            ORRRRR just get another box and put it in there!

                            Jeroen
                            Send message
                            Joined: Nov 25 05
                            Posts: 285
                            Credit: 294,883,253
                            RAC: 357,108
                            Message 110388 - Posted 11 Feb 2011 15:31:46 UTC - in response to Message 110361.

                              Just as an info, I run now 4 WUs at a time on a GTX 480 :

                              - GPU Load: +/- 84%
                              - Memory used: +/- 1350 Mb
                              - T°: 68°C
                              - Fan speed : 78%
                              - Run time: +/- 95 minutes

                              Cool :-)


                              95-minutes for 4 WUs is excellent. Nice work with that. :)

                              Profile [B^S] thierry@home
                              Avatar
                              Send message
                              Joined: Feb 19 05
                              Posts: 35
                              Credit: 1,250,712
                              RAC: 0
                              Message 110389 - Posted 11 Feb 2011 15:39:20 UTC - in response to Message 110388.

                                Last modified: 11 Feb 2011 15:39:35 UTC

                                yes, an extra 150 Mb of memory allows to run 5 WUs ;-)

                                The run time is 90 minutes in fact. 45 minutes if I run 1WU, 60 minutes for two WUs, 75 minutes for 3 WUs, ...
                                ____________


                                Intel I7 930 - GTX 480 - Windows 7 64

                                Join BOINC Synergy, the best team in the galaxy!

                                Profile astrocrab
                                Send message
                                Joined: Jan 28 08
                                Posts: 185
                                Credit: 65,716,770
                                RAC: 1,435
                                Message 110390 - Posted 11 Feb 2011 17:03:14 UTC - in response to Message 110389.

                                  same on 560:
                                  40m for 1 WU
                                  57m for 2 WU
                                  but i can't run 3 or more. :`((

                                  Profile [B^S] thierry@home
                                  Avatar
                                  Send message
                                  Joined: Feb 19 05
                                  Posts: 35
                                  Credit: 1,250,712
                                  RAC: 0
                                  Message 110396 - Posted 11 Feb 2011 18:09:20 UTC - in response to Message 110390.

                                    Last modified: 11 Feb 2011 18:23:59 UTC

                                    OK, ok don't make a nervous break down .... buy another GPU ;-)
                                    ____________


                                    Intel I7 930 - GTX 480 - Windows 7 64

                                    Join BOINC Synergy, the best team in the galaxy!

                                    Profile mickydl*
                                    Send message
                                    Joined: Oct 7 08
                                    Posts: 39
                                    Credit: 38,535,685
                                    RAC: 6,750
                                    Message 110397 - Posted 11 Feb 2011 18:17:40 UTC

                                      OK, here are some numbers for a GTX470 and Linux 64Bit:

                                      Load Time
                                      55% 0h55m for 1 WU
                                      79% 1h15m for 2 WU
                                      84% 1h55m for 3 WU

                                      The CPU is a Phenom IIx4 945 at stock speed. Running RNA on the free cores.

                                      Michael

                                      Profile [B^S] thierry@home
                                      Avatar
                                      Send message
                                      Joined: Feb 19 05
                                      Posts: 35
                                      Credit: 1,250,712
                                      RAC: 0
                                      Message 110400 - Posted 11 Feb 2011 19:36:59 UTC

                                        With this config, if I let Seti running, the GPU runs two Einstein and one Seti WU together.
                                        GPU Load: 90%
                                        Mem used: 1Gb
                                        ____________


                                        Intel I7 930 - GTX 480 - Windows 7 64

                                        Join BOINC Synergy, the best team in the galaxy!

                                        Profile astrocrab
                                        Send message
                                        Joined: Jan 28 08
                                        Posts: 185
                                        Credit: 65,716,770
                                        RAC: 1,435
                                        Message 110403 - Posted 11 Feb 2011 21:31:08 UTC - in response to Message 110397.

                                          OK, here are some numbers for a GTX470 and Linux 64Bit:
                                          Load Time
                                          79% 1h15m for 2 WU
                                          84% 1h55m for 3 WU
                                          Michael

                                          in this case there is no performance gain:
                                          2WU: 37.5 minutes for each
                                          3WU: 38.3 minutes for each.

                                          Profile astrocrab
                                          Send message
                                          Joined: Jan 28 08
                                          Posts: 185
                                          Credit: 65,716,770
                                          RAC: 1,435
                                          Message 110404 - Posted 11 Feb 2011 21:32:52 UTC - in response to Message 110400.

                                            With this config, if I let Seti running, the GPU runs two Einstein and one Seti WU together.
                                            GPU Load: 90%
                                            Mem used: 1Gb

                                            thank you, i will think about to run seti too )
                                            and maybe i will buy 560 with 2048M ram for experiments ))

                                            mikey
                                            Avatar
                                            Send message
                                            Joined: Jan 22 05
                                            Posts: 966
                                            Credit: 21,069,812
                                            RAC: 177,307
                                            Message 110420 - Posted 12 Feb 2011 10:36:50 UTC - in response to Message 110397.

                                              Last modified: 12 Feb 2011 10:37:20 UTC

                                              OK, here are some numbers for a GTX470 and Linux 64Bit:

                                              Load Time
                                              55% 0h55m for 1 WU
                                              79% 1h15m for 2 WU
                                              84% 1h55m for 3 WU

                                              The CPU is a Phenom IIx4 945 at stock speed. Running RNA on the free cores.

                                              Michael


                                              @mickydl* Where is your "CAL ATI Radeon HD5x00 series (Redwood) (1024MB) driver: 1.4.815" crunching? It seems that pc, AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2] (4 processors), is ONLY doing cpu units!

                                              Tony DeBari
                                              Send message
                                              Joined: Apr 29 05
                                              Posts: 30
                                              Credit: 11,024,503
                                              RAC: 5,907
                                              Message 110433 - Posted 12 Feb 2011 17:20:01 UTC - in response to Message 110420.

                                                OK, here are some numbers for a GTX470 and Linux 64Bit:

                                                Load Time
                                                55% 0h55m for 1 WU
                                                79% 1h15m for 2 WU
                                                84% 1h55m for 3 WU

                                                The CPU is a Phenom IIx4 945 at stock speed. Running RNA on the free cores.

                                                Michael


                                                @mickydl* Where is your "CAL ATI Radeon HD5x00 series (Redwood) (1024MB) driver: 1.4.815" crunching? It seems that pc, AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2] (4 processors), is ONLY doing cpu units!


                                                E@H doesn't yet have an app for ATI GPUs. Only CUDA GPUs are supported, and CUDA is an nVidia-only technology. I'm sure there are a lot of OpenCL-capable ATI/AMD cards out there that would love to get in on the action.


                                                -- Tony D.


                                                Profile mickydl*
                                                Send message
                                                Joined: Oct 7 08
                                                Posts: 39
                                                Credit: 38,535,685
                                                RAC: 6,750
                                                Message 110437 - Posted 12 Feb 2011 18:23:38 UTC - in response to Message 110420.

                                                  OK, here are some numbers for a GTX470 and Linux 64Bit:

                                                  Load Time
                                                  55% 0h55m for 1 WU
                                                  79% 1h15m for 2 WU
                                                  84% 1h55m for 3 WU

                                                  The CPU is a Phenom IIx4 945 at stock speed. Running RNA on the free cores.

                                                  Michael


                                                  @mickydl* Where is your "CAL ATI Radeon HD5x00 series (Redwood) (1024MB) driver: 1.4.815" crunching? It seems that pc, AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2] (4 processors), is ONLY doing cpu units!


                                                  There are not many projects that have ATI applications for LINUX. The only one that works without too much trouble is DNETC. That's what it is doing, though I would love to do something else with that card - like Einstein or Milkyway :)

                                                  Michael

                                                  mikey
                                                  Avatar
                                                  Send message
                                                  Joined: Jan 22 05
                                                  Posts: 966
                                                  Credit: 21,069,812
                                                  RAC: 177,307
                                                  Message 110459 - Posted 13 Feb 2011 10:44:13 UTC - in response to Message 110437.

                                                    OK, here are some numbers for a GTX470 and Linux 64Bit:

                                                    Load Time
                                                    55% 0h55m for 1 WU
                                                    79% 1h15m for 2 WU
                                                    84% 1h55m for 3 WU

                                                    The CPU is a Phenom IIx4 945 at stock speed. Running RNA on the free cores.

                                                    Michael


                                                    @mickydl* Where is your "CAL ATI Radeon HD5x00 series (Redwood) (1024MB) driver: 1.4.815" crunching? It seems that pc, AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2] (4 processors), is ONLY doing cpu units!


                                                    There are not many projects that have ATI applications for LINUX. The only one that works without too much trouble is DNETC. That's what it is doing, though I would love to do something else with that card - like Einstein or Milkyway :)

                                                    Michael


                                                    But your rac over there is only 20k, tou should be waaaaay above that with a 58xx card! I have a 5870 on Dnetc and am up above 260k and still climbing, I use Windows but that can't be the difference!
                                                    Here are the times and credits for one of my units:
                                                    904.91 11.95 3,091.20 DNETC@Home v1.31 (ati14)
                                                    and for one of your units:
                                                    10,777.46 99.83 3,292.45 DNETC@Home v1.02 (ati14)

                                                    For some reason yours is using a much different version of the Dnetc software, probably the Linux version. I can't believe it is that much slower!! My pc is doing units in about 15 minutes each, why is yours taking soooo much longer?!
                                                    ____________

                                                    Profile mickydl*
                                                    Send message
                                                    Joined: Oct 7 08
                                                    Posts: 39
                                                    Credit: 38,535,685
                                                    RAC: 6,750
                                                    Message 110468 - Posted 13 Feb 2011 12:51:52 UTC - in response to Message 110459.

                                                      I don't want to hijack this thread with an off-topic discussion so I'm moving it to ATI performance

                                                      Michael

                                                      Dirk
                                                      Send message
                                                      Joined: Jun 4 08
                                                      Posts: 35
                                                      Credit: 5,770,678
                                                      RAC: 10,572
                                                      Message 112385 - Posted 29 May 2011 23:11:03 UTC

                                                        Sorry for being ignorant, but I have a few questions.

                                                        I'd like to try and run more than one WU on my GTX 480 as well. Running just one seems awfully inefficient. From what I've read I think I'll need an XML file for that corresponding to the cuda WU I want to run. I guess it goes into einsteins project folder in my boinc data directory.

                                                        But do I just create a new XML document with something like the XML code posted earlier in this thread? I realize it'd need to have updated file versions and I'm also on win 7 64 bit instead of xp 32 bit.

                                                        Are there are pre-made XML files I could just use floating around or any how to guides? I tried to google but it's late and I'm sleepy and thus not finding anything useful.

                                                        Profile Gundolf Jahn
                                                        Send message
                                                        Joined: Mar 1 05
                                                        Posts: 1079
                                                        Credit: 324,682
                                                        RAC: 108
                                                        Message 112386 - Posted 30 May 2011 0:01:59 UTC - in response to Message 112385.

                                                          I tried to google but it's late and I'm sleepy and thus not finding anything useful.

                                                          You could google for "<app" or "app_info" (without the quotes) or try an advanced forum search (top left corner of this page).

                                                          But you should try to get some sleep beforehand. ;-)

                                                          Gruß,
                                                          Gundolf

                                                          mikey
                                                          Avatar
                                                          Send message
                                                          Joined: Jan 22 05
                                                          Posts: 966
                                                          Credit: 21,069,812
                                                          RAC: 177,307
                                                          Message 112396 - Posted 30 May 2011 13:49:53 UTC - in response to Message 112385.

                                                            Sorry for being ignorant, but I have a few questions.

                                                            I'd like to try and run more than one WU on my GTX 480 as well. Running just one seems awfully inefficient. From what I've read I think I'll need an XML file for that corresponding to the cuda WU I want to run. I guess it goes into einsteins project folder in my boinc data directory.

                                                            But do I just create a new XML document with something like the XML code posted earlier in this thread? I realize it'd need to have updated file versions and I'm also on win 7 64 bit instead of xp 32 bit.

                                                            Are there are pre-made XML files I could just use floating around or any how to guides? I tried to google but it's late and I'm sleepy and thus not finding anything useful.


                                                            I will throw something in the mix here, you need to see how much memory each current unit is using, then you can figure out if playing with an app_info.xml file will do any good. If for instance your current gpu has 512 meg of memory and an Einstein unit takes 350 meg then there is not enough space left over for another unit to run at the same time. If however you have a 2gb gpu and the same 350mb unit, then yes you can run multiple units on the same card.

                                                            To use multiple cards, which I think is what you are asking, you also need to check the current status of Boinc Manager and see what it says, if it says you are already running more than one gpu unit at a time, 4 in your case, than you may be maxed out depending on the memory requirements of each unit. If however you only see one unit running but have 4 gpu's in one machine then it may NOT an app_info.xml file you need but a cc_config.xml file which is entirely different. A line in the cc_config.xml file will ensure all gpu's on the machine are working, if all gpu's on the machine ARE working and crunching one unit, then you may not be able to separate them.

                                                            Dirk
                                                            Send message
                                                            Joined: Jun 4 08
                                                            Posts: 35
                                                            Credit: 5,770,678
                                                            RAC: 10,572
                                                            Message 112397 - Posted 30 May 2011 14:47:00 UTC - in response to Message 112396.

                                                              You probably looked at my old pc, it has an old 8800GT with 512 MB, my current pc has a GTX480 with 1500 MB. I'm not asking to use multiple cards btw, I only got 1 GPU per system, but I'd like that one GPU, at least the 480 to crunch more than one WU. If I can't it's much more productive for it to just run GPUGRID all the time but I like einstein@home.

                                                              Anyways, thanks for the tips, I found the stuff about the einstein apps in my client_state.xml. I'll need that for the app_info and then change a cuda setting from 1 to 0.25 right?

                                                              Any more advice is more than welcome though, a bit busy right now with other things so I'm waiting a bit before I start fiddling around with making an app_info.

                                                              FrankHagen
                                                              Send message
                                                              Joined: Feb 13 08
                                                              Posts: 102
                                                              Credit: 55,555
                                                              RAC: 0
                                                              Message 112398 - Posted 30 May 2011 15:06:34 UTC - in response to Message 112397.

                                                                Last modified: 30 May 2011 15:08:47 UTC

                                                                Any more advice is more than welcome though, a bit busy right now with other things so I'm waiting a bit before I start fiddling around with making an app_info.


                                                                spare at least 2 CPU-cores if you want to run 4 cuda-jobs at maximum speed.
                                                                don't start 4 of them at a time - give them an interleave of a minute.


                                                                sorry to say, but currently einstein has the worst performing cuda app around and it's a real pita to squeeze out a little bit on our side..

                                                                of course they could do what you want server side and give us an option to select the number of jobs we want to run in parallel. at least as long as the app is as it is now.

                                                                DanNeely
                                                                Send message
                                                                Joined: Sep 4 05
                                                                Posts: 1071
                                                                Credit: 56,505,635
                                                                RAC: 91,954
                                                                Message 112404 - Posted 31 May 2011 4:28:41 UTC - in response to Message 110377.

                                                                  I have an I7 CPU. It runs (for the moment):

                                                                  - 4 * Einstein (0,20 CPUs + 0,25 NVIDIA GPUs) +
                                                                  - 4 * Spinhenge WUs at 80% +
                                                                  - 4 * Spinhenge at 100%

                                                                  But the 0,20 CPUs, following the Tasks Manager, is in fact 0,40 CPUs.


                                                                  I think this is just a case of the estimate being for a nonHT cpu.
                                                                  ____________

                                                                  Profile Stranger7777
                                                                  Avatar
                                                                  Send message
                                                                  Joined: Mar 17 05
                                                                  Posts: 321
                                                                  Credit: 87,595,343
                                                                  RAC: 74,884
                                                                  Message 112407 - Posted 31 May 2011 8:11:08 UTC

                                                                    If anyone already made an app_info.xml with S6 WU's included, than it will be highly appreciated if one post it here.

                                                                    mikey
                                                                    Avatar
                                                                    Send message
                                                                    Joined: Jan 22 05
                                                                    Posts: 966
                                                                    Credit: 21,069,812
                                                                    RAC: 177,307
                                                                    Message 112410 - Posted 31 May 2011 11:27:29 UTC - in response to Message 112407.

                                                                      If anyone already made an app_info.xml with S6 WU's included, than it will be highly appreciated if one post it here.


                                                                      Both you and Dirk need to send a PM to Claggy and/or rroonnaalldd, they are both experts with app_info.xml files.

                                                                      Dirk how much memory does a use gpu workunit use right now? You will need twice that plus a little bit to fit more than one in memory, you need a little bit for the overhead.

                                                                      Dirk
                                                                      Send message
                                                                      Joined: Jun 4 08
                                                                      Posts: 35
                                                                      Credit: 5,770,678
                                                                      RAC: 10,572
                                                                      Message 112412 - Posted 31 May 2011 11:46:07 UTC - in response to Message 112410.

                                                                        Last modified: 31 May 2011 11:47:03 UTC

                                                                        If anyone already made an app_info.xml with S6 WU's included, than it will be highly appreciated if one post it here.


                                                                        Both you and Dirk need to send a PM to Claggy and/or rroonnaalldd, they are both experts with app_info.xml files.

                                                                        Dirk how much memory does a use gpu workunit use right now? You will need twice that plus a little bit to fit more than one in memory, you need a little bit for the overhead.


                                                                        100MB is used by windows it seems, 1 cuda WU from einstein adds around 300MB. So if I'd run 4 on my GTX480 it'd use roughly 1300 out of 1500MB. Edit, it'll probably be around the 1350MB mark just like for the op of this thread.



                                                                        I'll pm one of them soon, thanks for the heads up!

                                                                        Profile Shafa
                                                                        Send message
                                                                        Joined: May 31 05
                                                                        Posts: 22
                                                                        Credit: 77,320,144
                                                                        RAC: 145,953
                                                                        Message 112415 - Posted 31 May 2011 15:40:49 UTC

                                                                          Below, there is my app_info.xml for linux, running quite well.
                                                                          I am not sure if everything is correct, (too lazy to go through that carefully) but I have not experienced any trouble, yet.

                                                                          On these lines:
                                                                          <avg_ncpus>0.25</avg_ncpus>
                                                                          <max_ncpus>0.25</max_ncpus>
                                                                          should be value 0.2 - I am still playing with that number a bit

                                                                          I do not use binary radio pulsar search app. for CPU - only for GPU.

                                                                          1WU = approx. 320MB of graphics RAM
                                                                          2WU = approx. 570MB of graphics RAM
                                                                          3WU = approx. 860MB of graphics RAM (my system is not usable with 3WU - I have GTS250 with 1GB)


                                                                          <app_info>
                                                                          <app>
                                                                          <name>einstein_S5GC1HF</name>
                                                                          <user_friendly_name>Global Correlations S5 HF search #1</user_friendly_name>
                                                                          </app>
                                                                          <file_info>
                                                                          <name>einstein_S5GC1HF_1.07_i686-pc-linux-gnu__SSE2</name>
                                                                          <executable/>
                                                                          </file_info>
                                                                          <file_info>
                                                                          <name>einstein_S5R6_1.01_graphics_i686-pc-linux-gnu</name>
                                                                          <executable/>
                                                                          </file_info>
                                                                          <app_version>
                                                                          <app_name>einstein_S5GC1HF</app_name>
                                                                          <version_num>107</version_num>
                                                                          <platform>i686-pc-linux-gnu</platform>
                                                                          <avg_ncpus>1.000000</avg_ncpus>
                                                                          <max_ncpus>1.000000</max_ncpus>
                                                                          <plan_class>SSE2</plan_class>
                                                                          <file_ref>
                                                                          <file_name>einstein_S5GC1HF_1.07_i686-pc-linux-gnu__SSE2</file_name>
                                                                          <main_program/>
                                                                          </file_ref>
                                                                          <file_ref>
                                                                          <file_name>einstein_S5R6_1.01_graphics_i686-pc-linux-gnu</file_name>
                                                                          <open_name>graphics_app</open_name>
                                                                          </file_ref>
                                                                          </app_version>
                                                                          <app>
                                                                          <name>einstein_S6Bucket</name>
                                                                          <user_friendly_name>Gravitational Wave S6 GC search 1.01</user_friendly_name>
                                                                          </app>
                                                                          <file_info>
                                                                          <name>einstein_S6Bucket_1.01_i686-pc-linux-gnu__SSE2</name>
                                                                          <executable/>
                                                                          </file_info>
                                                                          <file_info>
                                                                          <name>einstein_S5R6_1.01_graphics_i686-pc-linux-gnu</name>
                                                                          <executable/>
                                                                          </file_info>
                                                                          <app_version>
                                                                          <app_name>einstein_S6Bucket</app_name>
                                                                          <version_num>101</version_num>
                                                                          <platform>i686-pc-linux-gnu</platform>
                                                                          <avg_ncpus>1.000000</avg_ncpus>
                                                                          <max_ncpus>1.000000</max_ncpus>
                                                                          <plan_class>SSE2</plan_class>
                                                                          <api_version>6.13.0</api_version>
                                                                          <file_ref>
                                                                          <file_name>einstein_S6Bucket_1.01_i686-pc-linux-gnu__SSE2</file_name>
                                                                          <main_program/>
                                                                          </file_ref>
                                                                          <file_ref>
                                                                          <file_name>einstein_S5R6_1.01_graphics_i686-pc-linux-gnu</file_name>
                                                                          <open_name>graphics_app</open_name>
                                                                          </file_ref>
                                                                          </app_version>
                                                                          <app>
                                                                          <name>einsteinbinary_BRP3</name>
                                                                          <user_friendly_name>Binary Radio Pulsar Search</user_friendly_name>
                                                                          </app>
                                                                          <file_info>
                                                                          <name>einsteinbinary_BRP3_1.08_i686-pc-linux-gnu__BRP3cuda32nv270</name>
                                                                          <executable/>
                                                                          </file_info>
                                                                          <file_info>
                                                                          <name>libcudart32_32_16.so</name>
                                                                          <executable/>
                                                                          </file_info>
                                                                          <file_info>
                                                                          <name>libcufft32_32_16.so</name>
                                                                          <executable/>
                                                                          </file_info>
                                                                          <file_info>
                                                                          <name>EULA.txt</name>
                                                                          </file_info>
                                                                          <file_info>
                                                                          <name>db.dev.lin.3d35195e</name>
                                                                          </file_info>
                                                                          <file_info>
                                                                          <name>dbhs.dev.lin.3d35195e</name>
                                                                          </file_info>
                                                                          <app_version>
                                                                          <app_name>einsteinbinary_BRP3</app_name>
                                                                          <version_num>108</version_num>
                                                                          <platform>i686-pc-linux-gnu</platform>
                                                                          <avg_ncpus>0.25</avg_ncpus>
                                                                          <max_ncpus>0.25</max_ncpus>
                                                                          <plan_class>BRP3cuda32nv270</plan_class>
                                                                          <api_version>6.13.0</api_version>
                                                                          <file_ref>
                                                                          <file_name>einsteinbinary_BRP3_1.08_i686-pc-linux-gnu__BRP3cuda32nv270</file_name>
                                                                          <main_program/>
                                                                          </file_ref>
                                                                          <file_ref>
                                                                          <file_name>libcudart32_32_16.so</file_name>
                                                                          <open_name>libcudart.so.3</open_name>
                                                                          <copy_file/>
                                                                          </file_ref>
                                                                          <file_ref>
                                                                          <file_name>libcufft32_32_16.so</file_name>
                                                                          <open_name>libcufft.so.3</open_name>
                                                                          <copy_file/>
                                                                          </file_ref>
                                                                          <file_ref>
                                                                          <file_name>EULA.txt</file_name>
                                                                          <open_name>EULA.txt</open_name>
                                                                          </file_ref>
                                                                          <file_ref>
                                                                          <file_name>db.dev.lin.3d35195e</file_name>
                                                                          <open_name>db.dev</open_name>
                                                                          <copy_file/>
                                                                          </file_ref>
                                                                          <file_ref>
                                                                          <file_name>dbhs.dev.lin.3d35195e</file_name>
                                                                          <open_name>dbhs.dev</open_name>
                                                                          <copy_file/>
                                                                          </file_ref>
                                                                          <coproc>
                                                                          <type>CUDA</type>
                                                                          <count>0.499995</count>
                                                                          </coproc>
                                                                          </app_version>
                                                                          </app_info>

                                                                          Claggy
                                                                          Send message
                                                                          Joined: Dec 29 06
                                                                          Posts: 391
                                                                          Credit: 902,099
                                                                          RAC: 2,556
                                                                          Message 112417 - Posted 31 May 2011 16:33:41 UTC - in response to Message 112410.

                                                                            Last modified: 31 May 2011 16:40:21 UTC

                                                                            Here's a complete app_info for windows with all the latest apps in it, just change the count value to 0.5, 0.33, or 0.25 depending if you want to run 2, 3 or 4 Cuda Wu's at once:

                                                                            <app_info>
                                                                            <app>
                                                                            <name>einstein_S5GC1HF</name>
                                                                            <user_friendly_name>Global Correlations S5 HF search #1</user_friendly_name>
                                                                            </app>
                                                                            <file_info>
                                                                            <name>einstein_S5GC1HF_3.06_windows_intelx86__S5GCESSE2.exe</name>
                                                                            <executable/>
                                                                            </file_info>
                                                                            <file_info>
                                                                            <name>einstein_S5R6_3.01_graphics_windows_intelx86.exe</name>
                                                                            <executable/>
                                                                            </file_info>

                                                                            <app_version>
                                                                            <app_name>einstein_S5GC1HF</app_name>
                                                                            <version_num>306</version_num>
                                                                            <platform>windows_intelx86</platform>
                                                                            <avg_ncpus>1.000000</avg_ncpus>
                                                                            <max_ncpus>1.000000</max_ncpus>
                                                                            <plan_class>S5GCESSE2</plan_class>
                                                                            <api_version>6.13.0</api_version>
                                                                            <file_ref>
                                                                            <file_name>einstein_S5GC1HF_3.06_windows_intelx86__S5GCESSE2.exe</file_name>
                                                                            <main_program/>
                                                                            </file_ref>
                                                                            <file_ref>
                                                                            <file_name>einstein_S5R6_3.01_graphics_windows_intelx86.exe</file_name>
                                                                            <open_name>graphics_app</open_name>
                                                                            </file_ref>
                                                                            </app_version>

                                                                            <app>
                                                                            <name>einsteinbinary_BRP3</name>
                                                                            <user_friendly_name>Binary Radio Pulsar Search</user_friendly_name>
                                                                            </app>
                                                                            <file_info>
                                                                            <name>einsteinbinary_BRP3_1.05_windows_intelx86__BRP3SSE.exe</name>
                                                                            <executable/>
                                                                            </file_info>
                                                                            <file_info>
                                                                            <name>einsteinbinary_BRP3_1.07_windows_intelx86__BRP3cuda32.exe</name>
                                                                            <executable/>
                                                                            </file_info>
                                                                            <file_info>
                                                                            <name>einsteinbinary_BRP3_1.00_graphics_windows_intelx86.exe</name>
                                                                            <executable/>
                                                                            </file_info>
                                                                            <file_info>
                                                                            <name>cudart_xp32_32_16.dll</name>
                                                                            <executable/>
                                                                            </file_info>
                                                                            <file_info>
                                                                            <name>cufft_xp32_32_16.dll</name>
                                                                            <executable/>
                                                                            </file_info>
                                                                            <file_info>
                                                                            <name>db.dev.win.3d35195e</name>
                                                                            </file_info>
                                                                            <file_info>
                                                                            <name>dbhs.dev.win.3d35195e</name>
                                                                            </file_info>

                                                                            <app_version>
                                                                            <app_name>einsteinbinary_BRP3</app_name>
                                                                            <version_num>105</version_num>
                                                                            <platform>windows_intelx86</platform>
                                                                            <avg_ncpus>1.000000</avg_ncpus>
                                                                            <max_ncpus>1.000000</max_ncpus>
                                                                            <plan_class>BRP3SSE</plan_class>
                                                                            <api_version>6.13.0</api_version>
                                                                            <file_ref>
                                                                            <file_name>einsteinbinary_BRP3_1.05_windows_intelx86__BRP3SSE.exe</file_name>
                                                                            <main_program/>
                                                                            </file_ref>
                                                                            <file_ref>
                                                                            <file_name>einsteinbinary_BRP3_1.00_graphics_windows_intelx86.exe</file_name>
                                                                            <open_name>graphics_app</open_name>
                                                                            </file_ref>
                                                                            </app_version>

                                                                            <app_version>
                                                                            <app_name>einsteinbinary_BRP3</app_name>
                                                                            <version_num>107</version_num>
                                                                            <platform>windows_intelx86</platform>
                                                                            <avg_ncpus>0.200000</avg_ncpus>
                                                                            <max_ncpus>1.000000</max_ncpus>
                                                                            <plan_class>BRP3cuda32</plan_class>
                                                                            <api_version>6.13.0</api_version>
                                                                            <file_ref>
                                                                            <file_name>einsteinbinary_BRP3_1.07_windows_intelx86__BRP3cuda32.exe</file_name>
                                                                            <main_program/>
                                                                            </file_ref>
                                                                            <file_ref>
                                                                            <file_name>cudart_xp32_32_16.dll</file_name>
                                                                            <open_name>cudart32_32_16.dll</open_name>
                                                                            <copy_file/>
                                                                            </file_ref>
                                                                            <file_ref>
                                                                            <file_name>cufft_xp32_32_16.dll</file_name>
                                                                            <open_name>cufft32_32_16.dll</open_name>
                                                                            <copy_file/>
                                                                            </file_ref>
                                                                            <file_ref>
                                                                            <file_name>einsteinbinary_BRP3_1.00_graphics_windows_intelx86.exe</file_name>
                                                                            <open_name>graphics_app</open_name>
                                                                            </file_ref>
                                                                            <file_ref>
                                                                            <file_name>db.dev.win.3d35195e</file_name>
                                                                            <open_name>db.dev</open_name>
                                                                            <copy_file/>
                                                                            </file_ref>
                                                                            <file_ref>
                                                                            <file_name>dbhs.dev.win.3d35195e</file_name>
                                                                            <open_name>dbhs.dev</open_name>
                                                                            <copy_file/>
                                                                            </file_ref>
                                                                            <coproc>
                                                                            <type>CUDA</type>
                                                                            <count>1.000000</count>
                                                                            </coproc>
                                                                            <gpu_ram>314572800.000000</gpu_ram>
                                                                            </app_version>

                                                                            <app>
                                                                            <name>einstein_S6Bucket</name>
                                                                            <user_friendly_name>Gravitational Wave S6 GC search</user_friendly_name>
                                                                            </app>
                                                                            <file_info>
                                                                            <name>einstein_S6Bucket_1.01_windows_intelx86__SSE2.exe</name>
                                                                            <executable/>
                                                                            </file_info>
                                                                            <file_info>
                                                                            <name>einstein_S5R6_3.01_graphics_windows_intelx86.exe</name>
                                                                            <executable/>
                                                                            </file_info>

                                                                            <app_version>
                                                                            <app_name>einstein_S6Bucket</app_name>
                                                                            <version_num>101</version_num>
                                                                            <platform>windows_intelx86</platform>
                                                                            <avg_ncpus>1.000000</avg_ncpus>
                                                                            <max_ncpus>1.000000</max_ncpus>
                                                                            <plan_class>SSE2</plan_class>
                                                                            <api_version>6.13.0</api_version>
                                                                            <file_ref>
                                                                            <file_name>einstein_S6Bucket_1.01_windows_intelx86__SSE2.exe</file_name>
                                                                            <main_program/>
                                                                            </file_ref>
                                                                            <file_ref>
                                                                            <file_name>einstein_S5R6_3.01_graphics_windows_intelx86.exe</file_name>
                                                                            <open_name>graphics_app</open_name>
                                                                            </file_ref>
                                                                            </app_version>
                                                                            </app_info>


                                                                            If you're missing apps and dll's, here are some links:

                                                                            http://einstein.aei.mpg.de/download/einstein_S5GC1HF_3.06_windows_intelx86__S5GCESSE2.exe
                                                                            http://einstein.aei.mpg.de/download/einstein_S5R6_3.01_graphics_windows_intelx86.exe
                                                                            http://einstein.aei.mpg.de/download/einsteinbinary_BRP3_1.05_windows_intelx86__BRP3SSE.exe
                                                                            http://einstein.aei.mpg.de/download/einsteinbinary_BRP3_1.07_windows_intelx86__BRP3cuda32.exe
                                                                            http://einstein.aei.mpg.de/download/einsteinbinary_BRP3_1.00_graphics_windows_intelx86.exe
                                                                            http://einstein.aei.mpg.de/download/cudart_xp32_32_16.dll
                                                                            http://einstein.aei.mpg.de/download/cufft_xp32_32_16.dll
                                                                            http://einstein.aei.mpg.de/download/db.dev.win.3d35195e
                                                                            http://einstein.aei.mpg.de/download/dbhs.dev.win.3d35195e
                                                                            http://einstein.aei.mpg.de/download/einstein_S6Bucket_1.01_windows_intelx86__SSE2.exe

                                                                            Claggy

                                                                            Dirk
                                                                            Send message
                                                                            Joined: Jun 4 08
                                                                            Posts: 35
                                                                            Credit: 5,770,678
                                                                            RAC: 10,572
                                                                            Message 112418 - Posted 31 May 2011 19:33:21 UTC - in response to Message 112417.

                                                                              Thanks so much Claggy! It works :)

                                                                              Running 2 WUs now. Also playing football manager so not gonna push it up to 3 or 4 for now.

                                                                              Dirk
                                                                              Send message
                                                                              Joined: Jun 4 08
                                                                              Posts: 35
                                                                              Credit: 5,770,678
                                                                              RAC: 10,572
                                                                              Message 112425 - Posted 1 Jun 2011 0:25:20 UTC

                                                                                Just a little update, running 4 WU's on the GPU now. RAM usage is fluctuating between 1200-1300 MB. But the core downclock bug kicked in with all the restarting of tasks so I'll restart now to see how stable it runs overnight.

                                                                                Restarted now. Running 6 CPU tasks along with the 4 GPU WUs. Overall CPU usage is at ~93% so the GPU isnt being starved of CPU cycles. GPU usage fluctuates around 73% (it was almost 100% when the core was downclocked), GPU memory use is between 1200-1300 MB. Here's a screenshot of MSI afterburners monitoring graphs.

                                                                                Profile Stranger7777
                                                                                Avatar
                                                                                Send message
                                                                                Joined: Mar 17 05
                                                                                Posts: 321
                                                                                Credit: 87,595,343
                                                                                RAC: 74,884
                                                                                Message 112429 - Posted 1 Jun 2011 6:00:19 UTC

                                                                                  Thanks Claggy! 10 minutes, every thing is allright.

                                                                                  Dirk
                                                                                  Send message
                                                                                  Joined: Jun 4 08
                                                                                  Posts: 35
                                                                                  Credit: 5,770,678
                                                                                  RAC: 10,572
                                                                                  Message 112432 - Posted 1 Jun 2011 6:44:19 UTC

                                                                                    Last modified: 1 Jun 2011 7:19:28 UTC

                                                                                    Hmm, I've run into a little problem. I can't get enough WUs in my queue. This is because apparently boinc thinks one cuda WU will now take me 34 hours to complete, despite evidence to the contrary. I remember reading somewhere how to change something that will help reduce that expected time but I just can't remember where.

                                                                                    I'd rather not increase my work buffer by too much so if anyone knows it's appreciated.

                                                                                    Edit:
                                                                                    Found it! Got the flops estimate from my client_state and added it to the app_info and changed the number by quite a bit.

                                                                                    <flops>16276995705.519375</flops>


                                                                                    Now the estimated runtime is 2 hours 40 mins which is good enough. I did manage to ruin my current batch of tasks though by fiddling around with the app_info, sorry for that! Also crashed my display driver when I exited boinc but I'm running the latest nvidia beta ones so it's not unexpected and it recovered just fine.

                                                                                    Last edit, all my latest tasks are called anonymous platform on my results page now. Is that normal? I don't think they were last night.

                                                                                    Profile Gundolf Jahn
                                                                                    Send message
                                                                                    Joined: Mar 1 05
                                                                                    Posts: 1079
                                                                                    Credit: 324,682
                                                                                    RAC: 108
                                                                                    Message 112433 - Posted 1 Jun 2011 7:44:11 UTC - in response to Message 112432.

                                                                                      Last modified: 1 Jun 2011 7:45:07 UTC

                                                                                      Last edit, all my latest tasks are called anonymous platform on my results page now. Is that normal? I don't think they were last night.

                                                                                      Yes, that's normal. By using an app_info.xml file, you tell the server that you are no longer using the project-supplied applications for the known platforms but an independent (anonymous) one.

                                                                                      That's also the drawback with going AP: you won't automatically get the newest applications but have to find and download them before incorporating them in your app_info.xml by hand.

                                                                                      Gruß,
                                                                                      Gundolf

                                                                                      Michael Karlinsky
                                                                                      Avatar
                                                                                      Send message
                                                                                      Joined: Jan 22 05
                                                                                      Posts: 894
                                                                                      Credit: 11,471,855
                                                                                      RAC: 13,709
                                                                                      Message 112435 - Posted 1 Jun 2011 8:17:49 UTC - in response to Message 112417.

                                                                                        Here's a complete app_info for windows with all the latest apps in it, just change the count value to 0.5, 0.33, or 0.25 depending if you want to run 2, 3 or 4 Cuda Wu's at once:

                                                                                        <snip>

                                                                                        Claggy


                                                                                        Hi,

                                                                                        why not create a project preference for that. Would make it a lot easier and less error prone.

                                                                                        Michael
                                                                                        ____________
                                                                                        Team Linux Users Everywhere

                                                                                        FrankHagen
                                                                                        Send message
                                                                                        Joined: Feb 13 08
                                                                                        Posts: 102
                                                                                        Credit: 55,555
                                                                                        RAC: 0
                                                                                        Message 112442 - Posted 1 Jun 2011 15:52:13 UTC - in response to Message 112425.

                                                                                          Just a little update, running 4 WU's on the GPU now. RAM usage is fluctuating between 1200-1300 MB. But the core downclock bug kicked in with all the restarting of tasks so I'll restart now to see how stable it runs overnight.

                                                                                          Restarted now. Running 6 CPU tasks along with the 4 GPU WUs. Overall CPU usage is at ~93% so the GPU isnt being starved of CPU cycles. GPU usage fluctuates around 73% (it was almost 100% when the core was downclocked), GPU memory use is between 1200-1300 MB.


                                                                                          if you can only reach about 75% gpu-load, the tasks are starving. don't care about overall CPU-usage - try to free another cpu-core and see if it improves..

                                                                                          Dirk
                                                                                          Send message
                                                                                          Joined: Jun 4 08
                                                                                          Posts: 35
                                                                                          Credit: 5,770,678
                                                                                          RAC: 10,572
                                                                                          Message 112444 - Posted 1 Jun 2011 17:02:44 UTC - in response to Message 112442.

                                                                                            Just a little update, running 4 WU's on the GPU now. RAM usage is fluctuating between 1200-1300 MB. But the core downclock bug kicked in with all the restarting of tasks so I'll restart now to see how stable it runs overnight.

                                                                                            Restarted now. Running 6 CPU tasks along with the 4 GPU WUs. Overall CPU usage is at ~93% so the GPU isnt being starved of CPU cycles. GPU usage fluctuates around 73% (it was almost 100% when the core was downclocked), GPU memory use is between 1200-1300 MB.


                                                                                            if you can only reach about 75% gpu-load, the tasks are starving. don't care about overall CPU-usage - try to free another cpu-core and see if it improves..


                                                                                            True, but even without running any cpu tasks the load only goes to 80%. I think it's because of win7. Got the same thing too on GPUGRID, it just maxes out at a set percentage (differs a bit per task there).

                                                                                            I think the best performance I get is when I leave 4 threads free for the GPU. Any more and it doesn't improve gpu usage. Even then each app only uses 5% of the CPU. Could that be changed by changing


                                                                                            <avg_ncpus>0.200000</avg_ncpus>


                                                                                            to something like?

                                                                                            <avg_ncpus>0.400000</avg_ncpus>

                                                                                            FrankHagen
                                                                                            Send message
                                                                                            Joined: Feb 13 08
                                                                                            Posts: 102
                                                                                            Credit: 55,555
                                                                                            RAC: 0
                                                                                            Message 112445 - Posted 1 Jun 2011 17:37:23 UTC - in response to Message 112444.

                                                                                              True, but even without running any cpu tasks the load only goes to 80%. I think it's because of win7. Got the same thing too on GPUGRID, it just maxes out at a set percentage (differs a bit per task there).


                                                                                              you got those nvidia-physics drivers installed? maybe those are slowing them down..

                                                                                              I think the best performance I get is when I leave 4 threads free for the GPU. Any more and it doesn't improve gpu usage. Even then each app only uses 5% of the CPU. Could that be changed by changing


                                                                                              <avg_ncpus>0.200000</avg_ncpus>


                                                                                              to something like?

                                                                                              <avg_ncpus>0.400000</avg_ncpus>


                                                                                              nope - this is only informative for boinc and does not change anything in the apps behaviour.

                                                                                              Dirk
                                                                                              Send message
                                                                                              Joined: Jun 4 08
                                                                                              Posts: 35
                                                                                              Credit: 5,770,678
                                                                                              RAC: 10,572
                                                                                              Message 112446 - Posted 1 Jun 2011 18:15:20 UTC

                                                                                                Maybe, but I also use this system for gaming and I'd rather keep physx for that. I haven't installed the 3D vision drivers though.

                                                                                                Sunny129
                                                                                                Avatar
                                                                                                Send message
                                                                                                Joined: Dec 5 05
                                                                                                Posts: 151
                                                                                                Credit: 92,449,852
                                                                                                RAC: 82,093
                                                                                                Message 112447 - Posted 1 Jun 2011 22:18:39 UTC - in response to Message 112446.

                                                                                                  Last modified: 1 Jun 2011 22:21:24 UTC

                                                                                                  Maybe, but I also use this system for gaming and I'd rather keep physx for that. I haven't installed the 3D vision drivers though.

                                                                                                  well at the very least you should try uninstalling the PhysX driver temporarily to see if in fact that is the reason your DC projects can't max out your GPU usage. at least then we'll know if PhysX had anything to do with it, and you can go right back to using PhysX. and on the off chance it turns out PhysX isn't the culprit, it would still be worth your while to explore why your GPU isn't reaching full usage under that kind of load...

                                                                                                  ...besides, inquiring minds would like to know :-)
                                                                                                  ____________

                                                                                                  Dirk
                                                                                                  Send message
                                                                                                  Joined: Jun 4 08
                                                                                                  Posts: 35
                                                                                                  Credit: 5,770,678
                                                                                                  RAC: 10,572
                                                                                                  Message 112450 - Posted 2 Jun 2011 9:19:35 UTC

                                                                                                    Well, it does reach full load in projects like milkyway and primegrid so I don't know.

                                                                                                    FrankHagen
                                                                                                    Send message
                                                                                                    Joined: Feb 13 08
                                                                                                    Posts: 102
                                                                                                    Credit: 55,555
                                                                                                    RAC: 0
                                                                                                    Message 112455 - Posted 2 Jun 2011 13:00:05 UTC - in response to Message 112450.

                                                                                                      Well, it does reach full load in projects like milkyway and primegrid so I don't know.


                                                                                                      maybe it's just a bandwidth limit on the bus. you may try to scale down to 3 tasks and see if it improves..

                                                                                                      Dirk
                                                                                                      Send message
                                                                                                      Joined: Jun 4 08
                                                                                                      Posts: 35
                                                                                                      Credit: 5,770,678
                                                                                                      RAC: 10,572
                                                                                                      Message 112457 - Posted 2 Jun 2011 17:55:42 UTC

                                                                                                        Last modified: 2 Jun 2011 18:02:02 UTC

                                                                                                        With 3 the load is lower. And doesn't PCIE x16 give plenty of bandwidth? It's not like I'm running 2 GPU's and the PCIE drops to 8x. I've been told on GPUGRID that win7 is about 11% slower than winXP and linux with their CUDA tasks because of the way win7 handles it. I think it's likely that's the cause here.

                                                                                                        1 WU is about 35%
                                                                                                        2 WU is about 60%
                                                                                                        3 WU is about 70%
                                                                                                        4 WU is about 80%

                                                                                                        It depends a bit on what else I'm doing with the PC or the kind of CPU tasks that Im running. But if I leave it alone 4 WU's running together will complete in about 90-100 minutes. I'm more than happy with this and it runs stable, what it doesn't seem to like is suspending all 4 cuda tasks at once though, that can sometimes cause a display driver crash and recovery (I'm thinking this may be because my GPU is overclocked by about 10%). Read somewhere it could be caused by a lack of power but my PSU is 850 watt which should be plenty for my system with just 1 GPU, and besides the GPU regularly gets stressed more while gaming (especially the witcher 2 but damn it looks awesome!).

                                                                                                        I'm hoping to get a new CPU cooler to replace the stock cooler in the next few weeks. I'll be able to overclock the CPU a bit then and maybe that will help the CUDA tasks out too.

                                                                                                        edit
                                                                                                        I also want to thank everyone for their feedback and advice, much appreciated!

                                                                                                        FrankHagen
                                                                                                        Send message
                                                                                                        Joined: Feb 13 08
                                                                                                        Posts: 102
                                                                                                        Credit: 55,555
                                                                                                        RAC: 0
                                                                                                        Message 112458 - Posted 2 Jun 2011 18:21:30 UTC - in response to Message 112457.

                                                                                                          With 3 the load is lower. And doesn't PCIE x16 give plenty of bandwidth? It's not like I'm running 2 GPU's and the PCIE drops to 8x. I've been told on GPUGRID that win7 is about 11% slower than winXP and linux with their CUDA tasks because of the way win7 handles it. I think it's likely that's the cause here.


                                                                                                          to my experience this is mostly due to all that nifty crap W7 has by default.

                                                                                                          as for einstein - yes, the linux-app running full CPU is a lot faster. on other projects like PG, collatz, GPUgrid it's just one or the other.

                                                                                                          It depends a bit on what else I'm doing with the PC or the kind of CPU tasks that Im running. But if I leave it alone 4 WU's running together will complete in about 90-100 minutes.


                                                                                                          that's probably as fast as you can get right now. of course it would be much better if the app would be improved and simply running a single WU would fully utilize the GPU.

                                                                                                          Read somewhere it could be caused by a lack of power but my PSU is 850 watt which should be plenty for my system with just 1 GPU, and besides the GPU regularly gets stressed more while gaming (especially the witcher 2 but damn it looks awesome!).


                                                                                                          that's silly! if the PSU would not be able to feed you host, you'd have real crashes.

                                                                                                          I also want to thank everyone for their feedback and advice, much appreciated!


                                                                                                          HTH!

                                                                                                          Profile John Jamulla
                                                                                                          Send message
                                                                                                          Joined: Feb 26 05
                                                                                                          Posts: 20
                                                                                                          Credit: 80,040,161
                                                                                                          RAC: 91,920
                                                                                                          Message 112967 - Posted 6 Jul 2011 22:08:25 UTC - in response to Message 110389.

                                                                                                            Hi - How do you find out your GPU load?
                                                                                                            I have a GTX 460 1GB videa card and I want to see how much it's being used.
                                                                                                            I also don't know how to change how many WU's it's doing at one time.
                                                                                                            I was trying to figure that out from your app_info file.

                                                                                                            Where is the manual/docs for the app_info file?

                                                                                                            I'm running windows 7 X64 on Intel Core 2 Q9300, 4 MB memory,
                                                                                                            ____________

                                                                                                            Richard Haselgrove
                                                                                                            Send message
                                                                                                            Joined: Dec 10 05
                                                                                                            Posts: 1304
                                                                                                            Credit: 28,178,310
                                                                                                            RAC: 11,835
                                                                                                            Message 112969 - Posted 6 Jul 2011 22:44:25 UTC - in response to Message 112967.

                                                                                                              Where is the manual/docs for the app_info file?

                                                                                                              Anonymous platform

                                                                                                              Dagorath
                                                                                                              Send message
                                                                                                              Joined: Apr 22 06
                                                                                                              Posts: 149
                                                                                                              Credit: 226,423
                                                                                                              RAC: 0
                                                                                                              Message 112970 - Posted 6 Jul 2011 22:58:32 UTC - in response to Message 112967.


                                                                                                                Where is the manual/docs for the app_info file?


                                                                                                                The app_info.xml file is part of the anonymous platform mechanism.

                                                                                                                ____________
                                                                                                                BOINC FAQ Service
                                                                                                                Official BOINC wiki
                                                                                                                Installing BOINC on Linux

                                                                                                                Profile Shafa
                                                                                                                Send message
                                                                                                                Joined: May 31 05
                                                                                                                Posts: 22
                                                                                                                Credit: 77,320,144
                                                                                                                RAC: 145,953
                                                                                                                Message 112971 - Posted 6 Jul 2011 23:19:58 UTC - in response to Message 112967.

                                                                                                                  Hi - How do you find out your GPU load?


                                                                                                                  run GPU Z, download it from HERE
                                                                                                                  ____________

                                                                                                                  Profile John Jamulla
                                                                                                                  Send message
                                                                                                                  Joined: Feb 26 05
                                                                                                                  Posts: 20
                                                                                                                  Credit: 80,040,161
                                                                                                                  RAC: 91,920
                                                                                                                  Message 112972 - Posted 7 Jul 2011 3:08:30 UTC - in response to Message 112971.

                                                                                                                    You guys are AWESOME! Thanks so much!!

                                                                                                                    Got the GPU-Z and GPU Monitor tools and can see what's going on now, though I don't really understand what's happening in a good vs. bad sense, "looks" good since it's running 3 GPU tasks instead of 1, but not sure how to tell until they finish if they are overall more throughput.
                                                                                                                    The GPU did seem to crash on me once so far, then went comp error one of the tasks.

                                                                                                                    FYI - I lost a lot of work when I did this fooling around with the app+info file, since I didn't know what I was doing. Sort of wiped out what I did and downloaded new stuff a bunch of times, but I think I have the hang of it as I'm changing things now in the app_info.xml file...

                                                                                                                    Why does BOINC delete a lot of the files/executables when it either doesn't think there's work etc.? Not sure why I keep having to copy stuff back. Not sure if it's because I have things in my app_info file that aren't currently being worked on??

                                                                                                                    So now what I have is: an Intel X9300 Quad core on GByte Mobo,at approx 2.5 Ghz, DDR2 PC2-6400 Mem (4 GBytes), NVIDIA GTX 460 1 GB overclocked quite a bit (ASUS card).

                                                                                                                    I am running 4 tasks on the CPU cores, and 3 cuda tasks on the single GTX 460. I have 1.05 and 1.07 BPS apps running simultaneously!! (.2 CPU + .33 NVIDIA GPU each says, 2 of the 3 are high priority.)

                                                                                                                    Just as an info, I run now 3 WUs at a time on a GTX 460, 4 on CPU cores:
                                                                                                                    I see all 4 CPU cores at 100%, and 2.2 of 4 GB Main memory being used.
                                                                                                                    GPU Monitor gives me temp of 97 deg F, GPU load of approx 92-98%, Vmem is approx 55-58%.
                                                                                                                    Ah! Using GPU-Z:
                                                                                                                    - GPU Load: +/- 91-99%
                                                                                                                    - Memory used: +/- 993 MB MAX
                                                                                                                    - Mem Ctrl: 59% MAX
                                                                                                                    - T°: 36°C
                                                                                                                    - Fan speed : 20%/1380 RPM


                                                                                                                    Should I be trying to get another task on the GPU to get it to 100% utilization?
                                                                                                                    I think the answer is NO, because not enough memory, right, already close to 100% GPU use as well?

                                                                                                                    How do I know if my CPUs being pegged are hurting feeding the GPU. I guess right now that, and knowing if it's truly faster are my issues??
                                                                                                                    I can wait to see how much work it's doing say tomorrow and how fast they're going.

                                                                                                                    Do you guys have some way you're benchmarking because just counting how many CUDA tasks finished in a certain amount of time manually?

                                                                                                                    I keep getting msgs about the one of the CUDA tasks not being able to run (hasn't really started yet) say every 30 secs or so that says not enough CUDU (3 others are already trying to run which I think is why). Not sure why it does this.


                                                                                                                    ____________

                                                                                                                    Profile Mike Hewson
                                                                                                                    Forum moderator
                                                                                                                    Avatar
                                                                                                                    Send message
                                                                                                                    Joined: Dec 1 05
                                                                                                                    Posts: 3322
                                                                                                                    Credit: 27,296,110
                                                                                                                    RAC: 20,198
                                                                                                                    Message 112973 - Posted 7 Jul 2011 3:42:32 UTC - in response to Message 112972.

                                                                                                                      Last modified: 7 Jul 2011 3:46:25 UTC

                                                                                                                      Why does BOINC delete a lot of the files/executables when it either doesn't think there's work etc.? Not sure why I keep having to copy stuff back. Not sure if it's because I have things in my app_info file that aren't currently being worked on??

                                                                                                                      This is 'locality scheduling' in action ( our server is chatting with your rig upon contact and deciding what to do next, preferably with what data files you already have ). See how it goes : I'd suggest it is just settling down with your new setup and BOINC specifically has no work ( project wide ) in the part of the parameter space that you held files for, alternatively it could be that others in the same part of the search space as yourself are chewing through quickly too .... Richard?

                                                                                                                      Should I be trying to get another task on the GPU to get it to 100% utilization? I think the answer is NO, because not enough memory, right, already close to 100% GPU use as well?

                                                                                                                      How do I know if my CPUs being pegged are hurting feeding the GPU. I guess right now that, and knowing if it's truly faster are my issues??

                                                                                                                      Your GPU is as fully utilised as possible, the CPUs are fine ( no harm ) and would be doing things other than feeding the GPU.

                                                                                                                      Do you guys have some way you're benchmarking because just counting how many CUDA tasks finished in a certain amount of time manually?

                                                                                                                      That's the way I do it, by checking the runtimes via BOINC 'Projects' tab -> select E@H -> hit 'Your computers' -> on the web page that appears in your browser select 'tasks' for the relevant rig -> the subsequent web page shows the 'Run time' ( wall clock ) and 'CPU time' ( thread time ).

                                                                                                                      I keep getting msgs about the one of the CUDA tasks not being able to run (hasn't really started yet) say every 30 secs or so that says not enough CUDU (3 others are already trying to run which I think is why). Not sure why it does this.

                                                                                                                      In the queue waiting their turn. After a while BOINC will adjust WU requests per your demonstrated load & this will settle down. You don't need to manage that.

                                                                                                                      Good to see you're having fun! :-)

                                                                                                                      Cheers, Mike.
                                                                                                                      ____________
                                                                                                                      "I have made this letter longer than usual, because I lack the time to make it short." - Blaise Pascal

                                                                                                                      Dirk
                                                                                                                      Send message
                                                                                                                      Joined: Jun 4 08
                                                                                                                      Posts: 35
                                                                                                                      Credit: 5,770,678
                                                                                                                      RAC: 10,572
                                                                                                                      Message 112975 - Posted 7 Jul 2011 6:10:02 UTC

                                                                                                                        Last modified: 7 Jul 2011 6:10:45 UTC

                                                                                                                        Oh nvm, it was F and not C. Ignore my post.

                                                                                                                        STE\/E [BlackOps]
                                                                                                                        Send message
                                                                                                                        Joined: Jan 18 05
                                                                                                                        Posts: 135
                                                                                                                        Credit: 5,921,526
                                                                                                                        RAC: 0
                                                                                                                        Message 113471 - Posted 8 Aug 2011 11:34:52 UTC

                                                                                                                          Last modified: 8 Aug 2011 11:41:32 UTC

                                                                                                                          I can't get any of the app's to work on my GTX 580 & Win XP to try and run 4 Wu's at once, all I keep getting is the following message:

                                                                                                                          [Einstein@Home] [error] State file error: missing application einsteinbinary_BRP4
                                                                                                                          [Einstein@Home] [error] Can't handle workunit in state file

                                                                                                                          Once I start BOINC the Einstein GPU Wu's get wiped out and no more is sent no matter how I configure the app file or what dll's I put in the Einstein project folder.

                                                                                                                          I've been trying to use the one claggy posted with no success:

                                                                                                                          EDIT: I didn't lose any Wu's though as I copied the entire BOINC Folder before messing around with an app file, I also disabled the Network activity for BOINC ... :)

                                                                                                                          <app_info>
                                                                                                                          <app>
                                                                                                                          <name>einstein_S5GC1HF</name>
                                                                                                                          <user_friendly_name>Global Correlations S5 HF search #1</user_friendly_name>
                                                                                                                          </app>
                                                                                                                          <file_info>
                                                                                                                          <name>einstein_S5GC1HF_3.06_windows_intelx86__S5GCESSE2.exe</name>
                                                                                                                          <executable/>
                                                                                                                          </file_info>
                                                                                                                          <file_info>
                                                                                                                          <name>einstein_S5R6_3.01_graphics_windows_intelx86.exe</name>
                                                                                                                          <executable/>
                                                                                                                          </file_info>

                                                                                                                          <app_version>
                                                                                                                          <app_name>einstein_S5GC1HF</app_name>
                                                                                                                          <version_num>306</version_num>
                                                                                                                          <platform>windows_intelx86</platform>
                                                                                                                          <avg_ncpus>1.000000</avg_ncpus>
                                                                                                                          <max_ncpus>1.000000</max_ncpus>
                                                                                                                          <plan_class>S5GCESSE2</plan_class>
                                                                                                                          <api_version>6.13.0</api_version>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>einstein_S5GC1HF_3.06_windows_intelx86__S5GCESSE2.exe</file_name>
                                                                                                                          <main_program/>
                                                                                                                          </file_ref>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>einstein_S5R6_3.01_graphics_windows_intelx86.exe</file_name>
                                                                                                                          <open_name>graphics_app</open_name>
                                                                                                                          </file_ref>
                                                                                                                          </app_version>

                                                                                                                          <app>
                                                                                                                          <name>einsteinbinary_BRP3</name>
                                                                                                                          <user_friendly_name>Binary Radio Pulsar Search</user_friendly_name>
                                                                                                                          </app>
                                                                                                                          <file_info>
                                                                                                                          <name>einsteinbinary_BRP3_1.05_windows_intelx86__BRP3SSE.exe</name>
                                                                                                                          <executable/>
                                                                                                                          </file_info>
                                                                                                                          <file_info>
                                                                                                                          <name>einsteinbinary_BRP3_1.07_windows_intelx86__BRP3cuda32.exe</name>
                                                                                                                          <executable/>
                                                                                                                          </file_info>
                                                                                                                          <file_info>
                                                                                                                          <name>einsteinbinary_BRP3_1.00_graphics_windows_intelx86.exe</name>
                                                                                                                          <executable/>
                                                                                                                          </file_info>
                                                                                                                          <file_info>
                                                                                                                          <name>cudart_xp32_32_16.dll</name>
                                                                                                                          <executable/>
                                                                                                                          </file_info>
                                                                                                                          <file_info>
                                                                                                                          <name>cufft_xp32_32_16.dll</name>
                                                                                                                          <executable/>
                                                                                                                          </file_info>
                                                                                                                          <file_info>
                                                                                                                          <name>db.dev.win.3d35195e</name>
                                                                                                                          </file_info>
                                                                                                                          <file_info>
                                                                                                                          <name>dbhs.dev.win.3d35195e</name>
                                                                                                                          </file_info>

                                                                                                                          <app_version>
                                                                                                                          <app_name>einsteinbinary_BRP3</app_name>
                                                                                                                          <version_num>105</version_num>
                                                                                                                          <platform>windows_intelx86</platform>
                                                                                                                          <avg_ncpus>1.000000</avg_ncpus>
                                                                                                                          <max_ncpus>1.000000</max_ncpus>
                                                                                                                          <plan_class>BRP3SSE</plan_class>
                                                                                                                          <api_version>6.13.0</api_version>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>einsteinbinary_BRP3_1.05_windows_intelx86__BRP3SSE.exe</file_name>
                                                                                                                          <main_program/>
                                                                                                                          </file_ref>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>einsteinbinary_BRP3_1.00_graphics_windows_intelx86.exe</file_name>
                                                                                                                          <open_name>graphics_app</open_name>
                                                                                                                          </file_ref>
                                                                                                                          </app_version>

                                                                                                                          <app_version>
                                                                                                                          <app_name>einsteinbinary_BRP3</app_name>
                                                                                                                          <version_num>107</version_num>
                                                                                                                          <platform>windows_intelx86</platform>
                                                                                                                          <avg_ncpus>0.200000</avg_ncpus>
                                                                                                                          <max_ncpus>1.000000</max_ncpus>
                                                                                                                          <plan_class>BRP3cuda32</plan_class>
                                                                                                                          <api_version>6.13.0</api_version>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>einsteinbinary_BRP3_1.07_windows_intelx86__BRP3cuda32.exe</file_name>
                                                                                                                          <main_program/>
                                                                                                                          </file_ref>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>cudart_xp32_32_16.dll</file_name>
                                                                                                                          <open_name>cudart32_32_16.dll</open_name>
                                                                                                                          <copy_file/>
                                                                                                                          </file_ref>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>cufft_xp32_32_16.dll</file_name>
                                                                                                                          <open_name>cufft32_32_16.dll</open_name>
                                                                                                                          <copy_file/>
                                                                                                                          </file_ref>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>einsteinbinary_BRP3_1.00_graphics_windows_intelx86.exe</file_name>
                                                                                                                          <open_name>graphics_app</open_name>
                                                                                                                          </file_ref>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>db.dev.win.3d35195e</file_name>
                                                                                                                          <open_name>db.dev</open_name>
                                                                                                                          <copy_file/>
                                                                                                                          </file_ref>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>dbhs.dev.win.3d35195e</file_name>
                                                                                                                          <open_name>dbhs.dev</open_name>
                                                                                                                          <copy_file/>
                                                                                                                          </file_ref>
                                                                                                                          <coproc>
                                                                                                                          <type>CUDA</type>
                                                                                                                          <count>1.000000</count>
                                                                                                                          </coproc>
                                                                                                                          <gpu_ram>314572800.000000</gpu_ram>
                                                                                                                          </app_version>

                                                                                                                          <app>
                                                                                                                          <name>einstein_S6Bucket</name>
                                                                                                                          <user_friendly_name>Gravitational Wave S6 GC search</user_friendly_name>
                                                                                                                          </app>
                                                                                                                          <file_info>
                                                                                                                          <name>einstein_S6Bucket_1.01_windows_intelx86__SSE2.exe</name>
                                                                                                                          <executable/>
                                                                                                                          </file_info>
                                                                                                                          <file_info>
                                                                                                                          <name>einstein_S5R6_3.01_graphics_windows_intelx86.exe</name>
                                                                                                                          <executable/>
                                                                                                                          </file_info>

                                                                                                                          <app_version>
                                                                                                                          <app_name>einstein_S6Bucket</app_name>
                                                                                                                          <version_num>101</version_num>
                                                                                                                          <platform>windows_intelx86</platform>
                                                                                                                          <avg_ncpus>1.000000</avg_ncpus>
                                                                                                                          <max_ncpus>1.000000</max_ncpus>
                                                                                                                          <plan_class>SSE2</plan_class>
                                                                                                                          <api_version>6.13.0</api_version>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>einstein_S6Bucket_1.01_windows_intelx86__SSE2.exe</file_name>
                                                                                                                          <main_program/>
                                                                                                                          </file_ref>
                                                                                                                          <file_ref>
                                                                                                                          <file_name>einstein_S5R6_3.01_graphics_windows_intelx86.exe</file_name>
                                                                                                                          <open_name>graphics_app</open_name>
                                                                                                                          </file_ref>
                                                                                                                          </app_version>
                                                                                                                          </app_info>

                                                                                                                          STE\/E [BlackOps]
                                                                                                                          Send message
                                                                                                                          Joined: Jan 18 05
                                                                                                                          Posts: 135
                                                                                                                          Credit: 5,921,526
                                                                                                                          RAC: 0
                                                                                                                          Message 113477 - Posted 8 Aug 2011 14:47:11 UTC

                                                                                                                            I finally got the app file figured out, running 6 @ a time on a Dual GTX 580 Setup, will have to see what the times are later ...

                                                                                                                            Profile Bikeman (Heinz-Bernd Eggenstein)
                                                                                                                            Forum moderator
                                                                                                                            Project administrator
                                                                                                                            Project developer
                                                                                                                            Avatar
                                                                                                                            Send message
                                                                                                                            Joined: Aug 28 06
                                                                                                                            Posts: 3195
                                                                                                                            Credit: 69,699,297
                                                                                                                            RAC: 59,194
                                                                                                                            Message 113481 - Posted 8 Aug 2011 17:41:07 UTC - in response to Message 113477.

                                                                                                                              Hi!


                                                                                                                              For those experimenting with multiple BRP app instances per GPU, it might be interesting to know that the "new" BRP4 units are quite a bit less CPU intensive than the formerly distributed BRP3 workunits. (while the app itself is the same, the signal data is different).

                                                                                                                              That means that the GPU load will be higher now, and the saving you get by running several units in parallel will be smaller. It will be interesting to see new runtime measurements.

                                                                                                                              HBE
                                                                                                                              ____________

                                                                                                                              joe areeda
                                                                                                                              Avatar
                                                                                                                              Send message
                                                                                                                              Joined: Dec 13 10
                                                                                                                              Posts: 286
                                                                                                                              Credit: 139,793,108
                                                                                                                              RAC: 44,993
                                                                                                                              Message 113482 - Posted 8 Aug 2011 17:59:33 UTC

                                                                                                                                Last modified: 8 Aug 2011 18:31:23 UTC

                                                                                                                                I've searched but am still clueless so please excuse the basic question.

                                                                                                                                I have a GTX560 w/1GB and would like to try 2 GPU apps at once. The system dual boots Win 7 Pro and Ubuntu 11.04.

                                                                                                                                I do not seem to have an app_info.xml file. Is there a writeup on how to create one from scratch? Are they universal enough that people are just copying the ones posted in this thread (and one other)?

                                                                                                                                Edit: Would a GT240 w/1GB benefit from 2 simultaneous apps?

                                                                                                                                Joe

                                                                                                                                Profile Bikeman (Heinz-Bernd Eggenstein)
                                                                                                                                Forum moderator
                                                                                                                                Project administrator
                                                                                                                                Project developer
                                                                                                                                Avatar
                                                                                                                                Send message
                                                                                                                                Joined: Aug 28 06
                                                                                                                                Posts: 3195
                                                                                                                                Credit: 69,699,297
                                                                                                                                RAC: 59,194
                                                                                                                                Message 113485 - Posted 8 Aug 2011 19:46:15 UTC

                                                                                                                                  Last modified: 8 Aug 2011 19:51:26 UTC

                                                                                                                                  Hi!

                                                                                                                                  App_info.xml files are specific to the operationg system and need to be changed whenever a new version of any of the apps listed in the file is released. The project as such does not generally recommend or offers support for writing app_info.xml files, but volunteers have posted files here.

                                                                                                                                  App_info.xml files require that you monitor your system and this forum (for new app releases) so in general I personally would only recommend it to expert users. Or put very bluntly: if you don't know how to write one, please think twice about whether you want to use one ;-)

                                                                                                                                  As for the GT240: with the new BRP4 workunits I would not expect a dramatic throughput increase by running two units in parallel. Maybe 10% ? It will be interesting to see the actual results.

                                                                                                                                  HBE
                                                                                                                                  ____________

                                                                                                                                  Richard Haselgrove
                                                                                                                                  Send message
                                                                                                                                  Joined: Dec 10 05
                                                                                                                                  Posts: 1304
                                                                                                                                  Credit: 28,178,310
                                                                                                                                  RAC: 11,835
                                                                                                                                  Message 113488 - Posted 8 Aug 2011 20:18:10 UTC - in response to Message 113485.

                                                                                                                                    As for the GT240: with the new BRP4 workunits I would not expect a dramatic throughput increase by running two units in parallel. Maybe 10% ? It will be interesting to see the actual results.

                                                                                                                                    And, to be very specific, the GT240 and all earlier-generation NVidia cards lack the context-switching hardware which make multiple-WU operation worthwhile on 4xx and 5xx series cards.

                                                                                                                                    In short, two apps will run simultaneously: but parallel running is likely to be less efficient than running in series, on older hardware.

                                                                                                                                    STE\/E [BlackOps]
                                                                                                                                    Send message
                                                                                                                                    Joined: Jan 18 05
                                                                                                                                    Posts: 135
                                                                                                                                    Credit: 5,921,526
                                                                                                                                    RAC: 0
                                                                                                                                    Message 113489 - Posted 8 Aug 2011 21:36:38 UTC

                                                                                                                                      I've run 1-2-3 & 4 at a time on each one of my dual GTX 580's and I don't see where I'm gaining anything. Running 1 @ a time is just as productive as running 2-3 or 4 as the times just go up 2-3 or 4 times the amount of time it takes to run 1 @ a time ... :/

                                                                                                                                      Jeroen
                                                                                                                                      Send message
                                                                                                                                      Joined: Nov 25 05
                                                                                                                                      Posts: 285
                                                                                                                                      Credit: 294,883,253
                                                                                                                                      RAC: 357,108
                                                                                                                                      Message 113490 - Posted 8 Aug 2011 22:03:43 UTC - in response to Message 113482.

                                                                                                                                        I've searched but am still clueless so please excuse the basic question.

                                                                                                                                        I have a GTX560 w/1GB and would like to try 2 GPU apps at once. The system dual boots Win 7 Pro and Ubuntu 11.04.

                                                                                                                                        I do not seem to have an app_info.xml file. Is there a writeup on how to create one from scratch? Are they universal enough that people are just copying the ones posted in this thread (and one other)?

                                                                                                                                        Edit: Would a GT240 w/1GB benefit from 2 simultaneous apps?

                                                                                                                                        Joe


                                                                                                                                        Here are links to both Linux and Windows app_info.xml files that I made for my systems. I've been able to see 8-13% performance increase running in Linux compared to Windows so Linux is probably the way to go.

                                                                                                                                        Linux App
                                                                                                                                        Win App

                                                                                                                                        The option for changing how many run at once is:

                                                                                                                                        <coproc>
                                                                                                                                        <type>CUDA</type>
                                                                                                                                        <count>0.500000</count>
                                                                                                                                        </coproc>

                                                                                                                                        1.0 for one unit per GPU, .50 for two units per GPU, .33 for three units per GPU, etc.

                                                                                                                                        The files still have BRP3 GPU in them as I still get a few of those from time-to-time. You also have the option of removing the CPU related sections if you plan to run GPU only to simplify the configuration file. You have to make sure all the necessary project files as specified in the XML are available before using the XML file.

                                                                                                                                        STE\/E [BlackOps]
                                                                                                                                        Send message
                                                                                                                                        Joined: Jan 18 05
                                                                                                                                        Posts: 135
                                                                                                                                        Credit: 5,921,526
                                                                                                                                        RAC: 0
                                                                                                                                        Message 113491 - Posted 9 Aug 2011 19:54:20 UTC - in response to Message 113488.

                                                                                                                                          As for the GT240: with the new BRP4 workunits I would not expect a dramatic throughput increase by running two units in parallel. Maybe 10% ? It will be interesting to see the actual results.

                                                                                                                                          And, to be very specific, the GT240 and all earlier-generation NVidia cards lack the context-switching hardware which make multiple-WU operation worthwhile on 4xx and 5xx series cards.

                                                                                                                                          In short, two apps will run simultaneously: but parallel running is likely to be less efficient than running in series, on older hardware.


                                                                                                                                          I think running 3 (Memory Permitting) is the most Optimal, anything over that & the run times start increasing Per Wu ... Just observation running different amounts on several GTX 580 Box's ...

                                                                                                                                          Profile Bikeman (Heinz-Bernd Eggenstein)
                                                                                                                                          Forum moderator
                                                                                                                                          Project administrator
                                                                                                                                          Project developer
                                                                                                                                          Avatar
                                                                                                                                          Send message
                                                                                                                                          Joined: Aug 28 06
                                                                                                                                          Posts: 3195
                                                                                                                                          Credit: 69,699,297
                                                                                                                                          RAC: 59,194
                                                                                                                                          Message 113492 - Posted 9 Aug 2011 20:15:12 UTC - in response to Message 113491.

                                                                                                                                            As for the GT240: with the new BRP4 workunits I would not expect a dramatic throughput increase by running two units in parallel. Maybe 10% ? It will be interesting to see the actual results.

                                                                                                                                            And, to be very specific, the GT240 and all earlier-generation NVidia cards lack the context-switching hardware which make multiple-WU operation worthwhile on 4xx and 5xx series cards.

                                                                                                                                            In short, two apps will run simultaneously: but parallel running is likely to be less efficient than running in series, on older hardware.


                                                                                                                                            I think running 3 (Memory Permitting) is the most Optimal, anything over that & the run times start increasing Per Wu ... Just observation running different amounts on several GTX 580 Box's ...



                                                                                                                                            Wow, you have an impressive array of hosts there!!! Most of your hosts even come with 2 cards !?! But you were not running 2 x 3 WU in parallel, (3 on each GPU), right??

                                                                                                                                            HB



                                                                                                                                            ____________

                                                                                                                                            STE\/E [BlackOps]
                                                                                                                                            Send message
                                                                                                                                            Joined: Jan 18 05
                                                                                                                                            Posts: 135
                                                                                                                                            Credit: 5,921,526
                                                                                                                                            RAC: 0
                                                                                                                                            Message 113494 - Posted 9 Aug 2011 20:26:29 UTC - in response to Message 113492.

                                                                                                                                              As for the GT240: with the new BRP4 workunits I would not expect a dramatic throughput increase by running two units in parallel. Maybe 10% ? It will be interesting to see the actual results.

                                                                                                                                              And, to be very specific, the GT240 and all earlier-generation NVidia cards lack the context-switching hardware which make multiple-WU operation worthwhile on 4xx and 5xx series cards.

                                                                                                                                              In short, two apps will run simultaneously: but parallel running is likely to be less efficient than running in series, on older hardware.


                                                                                                                                              I think running 3 (Memory Permitting) is the most Optimal, anything over that & the run times start increasing Per Wu ... Just observation running different amounts on several GTX 580 Box's ...



                                                                                                                                              Wow, you have an impressive array of hosts there!!! Most of your hosts even come with 2 cards !?! But you were not running 2 x 3 WU in parallel, (3 on each GPU), right??

                                                                                                                                              HB




                                                                                                                                              Right, I was only running 1 on each GPU until a few days ago. I couldn't get an app file to work until then, now I'm just running 2 Dual 580 Box's & settled in on 3 Per GPU as by my figures that's the most Optimal on the GTX 580's I have ...

                                                                                                                                              joe areeda
                                                                                                                                              Avatar
                                                                                                                                              Send message
                                                                                                                                              Joined: Dec 13 10
                                                                                                                                              Posts: 286
                                                                                                                                              Credit: 139,793,108
                                                                                                                                              RAC: 44,993
                                                                                                                                              Message 113499 - Posted 10 Aug 2011 2:04:48 UTC - in response to Message 113485.

                                                                                                                                                App_info.xml files require that you monitor your system and this forum (for new app releases) so in general I personally would only recommend it to expert users. Or put very bluntly: if you don't know how to write one, please think twice about whether you want to use one ;-)

                                                                                                                                                I appreciate the honesty (bluntness) I sometimes miss the subtleties.

                                                                                                                                                I consider myself trainable and since I'm in the explore and learn stage I don't mind getting into a little trouble.

                                                                                                                                                If I understand properly how this works copying the appropriate file Jeroen linked to into my BOINC directory and changing the coprocessor count 0.5 should allow 2 CUDA tasks to run.

                                                                                                                                                The problem will be I have to watch for new versions and modify the file myself, or delete it and go back to one CUDA task at a time. If I miss an update the problem would be no tasks for the new program. Right?

                                                                                                                                                Right now it looks like I am using 492 our of 1024 MB of GPU memory with 60-70% or the processors resulting in one BPR4 task every 35 min or so. Temps are pretty steady at 59C with the fan at 40%. Looks like there is excess capacity.

                                                                                                                                                This is my new home machine so I keep a pretty close eye on it. I may upgrade that 240 and give it to my son.

                                                                                                                                                Joe

                                                                                                                                                Profile Gundolf Jahn
                                                                                                                                                Send message
                                                                                                                                                Joined: Mar 1 05
                                                                                                                                                Posts: 1079
                                                                                                                                                Credit: 324,682
                                                                                                                                                RAC: 108
                                                                                                                                                Message 113500 - Posted 10 Aug 2011 7:14:33 UTC - in response to Message 113499.

                                                                                                                                                  Last modified: 10 Aug 2011 7:15:09 UTC

                                                                                                                                                  If I understand properly how this works copying the appropriate file Jeroen linked to into my BOINC directory and changing the coprocessor count 0.5 should allow 2 CUDA tasks to run.

                                                                                                                                                  Not quite. The app_info.xml file (and all other files mentioned therein) go(es) to the appropriate project directory (einstein.phys.uwm.edu in this case), which is located in the projects subdirectory of the BOINC data directory.

                                                                                                                                                  Gruß,
                                                                                                                                                  Gundolf

                                                                                                                                                  joe areeda
                                                                                                                                                  Avatar
                                                                                                                                                  Send message
                                                                                                                                                  Joined: Dec 13 10
                                                                                                                                                  Posts: 286
                                                                                                                                                  Credit: 139,793,108
                                                                                                                                                  RAC: 44,993
                                                                                                                                                  Message 113510 - Posted 10 Aug 2011 12:55:27 UTC - in response to Message 113500.

                                                                                                                                                    If I understand properly how this works copying the appropriate file Jeroen linked to into my BOINC directory and changing the coprocessor count 0.5 should allow 2 CUDA tasks to run.

                                                                                                                                                    Not quite. The app_info.xml file (and all other files mentioned therein) go(es) to the appropriate project directory (einstein.phys.uwm.edu in this case), which is located in the projects subdirectory of the BOINC data directory.

                                                                                                                                                    Gruß,
                                                                                                                                                    Gundolf

                                                                                                                                                    Thanks,
                                                                                                                                                    I'll give it a try to see what happens. The more I think about the more I think I'll probably run without it in the long term.

                                                                                                                                                    This system has only been up for a few days and hasn't got a stable RAC yet but it looks like almost 20K credits/day, no overclocking. I'm guessing another GPU task will add 2 or 3K.

                                                                                                                                                    Joe


                                                                                                                                                    Thomas
                                                                                                                                                    Send message
                                                                                                                                                    Joined: Aug 27 11
                                                                                                                                                    Posts: 7
                                                                                                                                                    Credit: 7,152,760
                                                                                                                                                    RAC: 0
                                                                                                                                                    Message 114101 - Posted 18 Sep 2011 10:35:44 UTC

                                                                                                                                                      Last modified: 18 Sep 2011 10:50:43 UTC

                                                                                                                                                      Would be there a real gain on BRP4 CUDA tasks with GTX260?

                                                                                                                                                      Highlander
                                                                                                                                                      Send message
                                                                                                                                                      Joined: Jul 1 05
                                                                                                                                                      Posts: 12
                                                                                                                                                      Credit: 20,086,979
                                                                                                                                                      RAC: 28,031
                                                                                                                                                      Message 114112 - Posted 18 Sep 2011 20:10:24 UTC - in response to Message 114101.

                                                                                                                                                        Would be there a real gain on BRP4 CUDA tasks with GTX260?


                                                                                                                                                        no, only on Fermi like GPUs. (there was a better performance with BRP3 tasks, but not with BRP4)
                                                                                                                                                        ____________

                                                                                                                                                        uBronan
                                                                                                                                                        Avatar
                                                                                                                                                        Send message
                                                                                                                                                        Joined: May 28 07
                                                                                                                                                        Posts: 10
                                                                                                                                                        Credit: 231,037
                                                                                                                                                        RAC: 0
                                                                                                                                                        Message 114169 - Posted 23 Sep 2011 13:34:11 UTC

                                                                                                                                                          I still think running more then 2 instances on any gpu is not worth doing.
                                                                                                                                                          The only gain you have with running 3 is that it actually do the first 2 and when the gpu has some free time spends that on the 3th unit.
                                                                                                                                                          But basically it can only do 2 as far as i know nothing has been changed in the current hardware to support more then those 2 instances.
                                                                                                                                                          It does not matter if its a ati/nvidia card or even if its a gts250 as long as it has enough video memory and is not fully loaded you can add a second process.
                                                                                                                                                          Even my little GT440 with 512 Mb gddr5 can do 2 instances, nevertheless this ofcourse is not worthwhile doing
                                                                                                                                                          Only the medium and fast cards really benefit from 2 instances.
                                                                                                                                                          As Steve already said the gains can be minimal on projects if the project needs much contact with the cpu like einstein
                                                                                                                                                          Projects like milkyway, primegrid and collatz show a much bigger gain cause they hardly use cpu, they simply give the calculation to the gpu with some checks on how far it is.
                                                                                                                                                          Else they hardly have to interfere with the process as soon as it has been feeded to the gpu. Milkyway even crashes if you stop it while running.
                                                                                                                                                          It simply needs to end the work before control returns to the cpu.


                                                                                                                                                          ____________

                                                                                                                                                          duftkerze
                                                                                                                                                          Send message
                                                                                                                                                          Joined: Jun 10 05
                                                                                                                                                          Posts: 1
                                                                                                                                                          Credit: 3,269,987
                                                                                                                                                          RAC: 5,919
                                                                                                                                                          Message 114432 - Posted 9 Oct 2011 12:39:26 UTC

                                                                                                                                                            its my app_info:

                                                                                                                                                            <app_info>
                                                                                                                                                            <app>
                                                                                                                                                            <name>einsteinbinary_BRP4</name>
                                                                                                                                                            <user_friendly_name>Binary Radio Pulsar Search</user_friendly_name>
                                                                                                                                                            </app>
                                                                                                                                                            <file_info>
                                                                                                                                                            <name>einsteinbinary_BRP4_1.00_windows_intelx86__BRP3cuda32.exe</name>
                                                                                                                                                            <executable></executable>
                                                                                                                                                            </file_info>
                                                                                                                                                            <file_info>
                                                                                                                                                            <name>cudart_xp32_32_16.dll</name>
                                                                                                                                                            <executable></executable>
                                                                                                                                                            </file_info>
                                                                                                                                                            <file_info>
                                                                                                                                                            <name>cufft_xp32_32_16.dll</name>
                                                                                                                                                            <executable></executable>
                                                                                                                                                            </file_info>
                                                                                                                                                            <file_info>
                                                                                                                                                            <name>db.dev.win.3d35195e</name>
                                                                                                                                                            </file_info>
                                                                                                                                                            <file_info>
                                                                                                                                                            <name>dbhs.dev.win.3d35195e</name>
                                                                                                                                                            </file_info>
                                                                                                                                                            <app_version>
                                                                                                                                                            <app_name>einsteinbinary_BRP4</app_name>
                                                                                                                                                            <version_num>100</version_num>
                                                                                                                                                            <platform>windows_x86_64</platform>
                                                                                                                                                            <avg_ncpus>0.200000</avg_ncpus>
                                                                                                                                                            <max_ncpus>1.000000</max_ncpus>
                                                                                                                                                            <plan_class>BRP3cuda32</plan_class>
                                                                                                                                                            <api_version>6.13.0</api_version>
                                                                                                                                                            <file_ref>
                                                                                                                                                            <file_name>einsteinbinary_BRP4_1.00_windows_intelx86__BRP3cuda32.exe</file_name>
                                                                                                                                                            <main_program></main_program>
                                                                                                                                                            </file_ref>
                                                                                                                                                            <file_ref>
                                                                                                                                                            <file_name>cudart_xp32_32_16.dll</file_name>
                                                                                                                                                            <open_name>cudart32_32_16.dll</open_name>
                                                                                                                                                            <copy_file></copy_file>
                                                                                                                                                            </file_ref>
                                                                                                                                                            <file_ref>
                                                                                                                                                            <file_name>cufft_xp32_32_16.dll</file_name>
                                                                                                                                                            <open_name>cufft32_32_16.dll</open_name>
                                                                                                                                                            <copy_file></copy_file>
                                                                                                                                                            </file_ref>
                                                                                                                                                            <file_ref>
                                                                                                                                                            <file_name>db.dev.win.3d35195e</file_name>
                                                                                                                                                            <open_name>db.dev</open_name>
                                                                                                                                                            <copy_file></copy_file>
                                                                                                                                                            </file_ref>
                                                                                                                                                            <file_ref>
                                                                                                                                                            <file_name>dbhs.dev.win.3d35195e</file_name>
                                                                                                                                                            <open_name>dbhs.dev</open_name>
                                                                                                                                                            <copy_file></copy_file>
                                                                                                                                                            </file_ref>
                                                                                                                                                            <coproc>
                                                                                                                                                            <type>CUDA</type>
                                                                                                                                                            <count>0.5</count>
                                                                                                                                                            </coproc>
                                                                                                                                                            <file_ref>
                                                                                                                                                            <file_name>einsteinbinary_BRP4_1.00_windows_intelx86__BRP3cuda32.exe</file_name>
                                                                                                                                                            <main_program></main_program>
                                                                                                                                                            </file_ref>
                                                                                                                                                            <gpu_ram>314572800.000000</gpu_ram>
                                                                                                                                                            </app_version>
                                                                                                                                                            </app_info>

                                                                                                                                                            whats false?


                                                                                                                                                            failure: couldn't start No main program specified



                                                                                                                                                            ____________

                                                                                                                                                            Richard Haselgrove
                                                                                                                                                            Send message
                                                                                                                                                            Joined: Dec 10 05
                                                                                                                                                            Posts: 1304
                                                                                                                                                            Credit: 28,178,310
                                                                                                                                                            RAC: 11,835
                                                                                                                                                            Message 114433 - Posted 9 Oct 2011 13:04:58 UTC - in response to Message 114432.

                                                                                                                                                              failure: couldn't start No main program specified

                                                                                                                                                              Refer to the app_info documentation

                                                                                                                                                              The format for directives like that is

                                                                                                                                                              <main_program/>

                                                                                                                                                              - a self-closing tag on its own, not an empty open/close pair.

                                                                                                                                                              Likewise <executable/>, <copy_file/>

                                                                                                                                                              Sid
                                                                                                                                                              Send message
                                                                                                                                                              Joined: Oct 17 10
                                                                                                                                                              Posts: 87
                                                                                                                                                              Credit: 43,499,579
                                                                                                                                                              RAC: 67,888
                                                                                                                                                              Message 114434 - Posted 9 Oct 2011 13:15:26 UTC - in response to Message 114169.

                                                                                                                                                                I still think running more then 2 instances on any gpu is not worth doing.

                                                                                                                                                                Probably, it is right for more slow cards but just some statistics from GTX 560 Ti card with 2Gb memory:
                                                                                                                                                                1. only one BRP4 Wu - about 38 minutes
                                                                                                                                                                2. 6 BRP4 Wus samulteniously - about 1 hours and 30 minutes so 150/6= 25 minutes for one Wu.

                                                                                                                                                                Sid
                                                                                                                                                                Send message
                                                                                                                                                                Joined: Oct 17 10
                                                                                                                                                                Posts: 87
                                                                                                                                                                Credit: 43,499,579
                                                                                                                                                                RAC: 67,888
                                                                                                                                                                Message 114438 - Posted 9 Oct 2011 16:39:28 UTC - in response to Message 114434.

                                                                                                                                                                  I still think running more then 2 instances on any gpu is not worth doing.


                                                                                                                                                                  2. 6 BRP4 Wus samulteniously - about 1 hours and 30 minutes so 150/6= 25 minutes for one Wu.


                                                                                                                                                                  Sorry, TWO hours and 30 minutes.

                                                                                                                                                                  Fred J. Verster
                                                                                                                                                                  Avatar
                                                                                                                                                                  Send message
                                                                                                                                                                  Joined: Apr 27 08
                                                                                                                                                                  Posts: 114
                                                                                                                                                                  Credit: 20,727,834
                                                                                                                                                                  RAC: 0
                                                                                                                                                                  Message 114449 - Posted 11 Oct 2011 19:57:18 UTC - in response to Message 114438.

                                                                                                                                                                    Last modified: 11 Oct 2011 20:32:52 UTC

                                                                                                                                                                    [ADDED]
                                                                                                                                                                    2 hours 30 is 9000 seconds, how much time takes 1, at a time?

                                                                                                                                                                    I'm running a Q6600+GTX470 & a X9650(@3.51GHz)+GTX480.
                                                                                                                                                                    1 WU on each GPU, 2000 seconds on the 480 and ~2500 seconds
                                                                                                                                                                    on the 470.
                                                                                                                                                                    Since the 480 has 1560 MByte DDR5 and the 470 1280MByte RAM, so 2
                                                                                                                                                                    is an option, IMHO, if compute time doesn't increase, too much!

                                                                                                                                                                    I did run 3 SETI MB WUs on the 480, changed that to 2, cause of lack
                                                                                                                                                                    of work and too much heat. (It now runs 'case-less' a 'modular- build')
                                                                                                                                                                    ____________

                                                                                                                                                                    Knight who says Ni N! N!

                                                                                                                                                                    archae86
                                                                                                                                                                    Send message
                                                                                                                                                                    Joined: Dec 6 05
                                                                                                                                                                    Posts: 970
                                                                                                                                                                    Credit: 87,684,959
                                                                                                                                                                    RAC: 145,219
                                                                                                                                                                    Message 114491 - Posted 14 Oct 2011 3:28:55 UTC - in response to Message 113481.

                                                                                                                                                                      Bikeman wrote:
                                                                                                                                                                      the "new" BRP4 units are quite a bit less CPU intensive than the formerly distributed BRP3 workunits. (while the app itself is the same, the signal data is different).

                                                                                                                                                                      That means that the GPU load will be higher now, and the saving you get by running several units in parallel will be smaller. It will be interesting to see new runtime measurements.
                                                                                                                                                                      I've been working with my new host which has a GTX 460 graphics card of the Gigabyte SOC flavor.

                                                                                                                                                                      Single WU at a time gives a very tight distribution of elapsed times averaging 1900 seconds, with a stdev of something like 20 seconds. GPU load generally was just under 70%.

                                                                                                                                                                      But running two WU at a time exhibited bimodal behavior. Much of the time the system was running at little if any higher throughput than single WU (as shown by all of rate of progress, GPU load, and power consumption) but at times it would run materially faster (again as manifested by all three). About 50 WUs processed over about a day showed an average throughput advantage over single WU of a bit over 8%.

                                                                                                                                                                      When I first started running three simultaneous WU, all three indicators suggested considerable further improvement. In particular GPU load was mostly about 88%. One of the three active WUs progressed much faster than the other two. But as soon as it finished, this desirable behavior vanished, and since then I've seen GPU load at 77%, power consumption to match, and matched progress on all three WUs with throughput no higher than the 2 WU case average--possibly a bit lower.

                                                                                                                                                                      An additional problem is that the server is generally only awarding this host one BRP4 WU for each request. Sometimes a second or third request is generated and award of one WU each granted at one minute intervals, but then the four hour delay penalty for use of anonymous platform is posted. As the host consumes about eight WUs in four hours if available, this is a problem for unattended operation. When I was trying two WU operation yesterday, the host was given enough work to stay busy--I don't know what the difference is today.

                                                                                                                                                                      In all three conditions, the forecast run time has not converged toward the real one on any reasonable time scale (well past the oft-mentioned 10 returned WU point). For single WU work (with no app_info.xml file) after many dozens of results returned, the estimate was about double truth. At double WU running, the error was much larger, though I failed to log the values. Now at triple WU running, for which the real elapsed time for each WU is about an hour and a half, the initial estimate was over 21 hours, and has so far declined only to 19.5 hours. As I've requested a 3.5 day queue, this alone, however, seems not enough reason for the parsimonious distribution of work.

                                                                                                                                                                      While both double and triple WU operation has given a modest performance boost, unless I can get the higher activity condition observed on some work to be typical by some adjustment, the improvement seems not worth the overhead and risk associated with anonymous platform operation.

                                                                                                                                                                      ____________

                                                                                                                                                                      Post to thread

                                                                                                                                                                      Message boards : Cruncher's Corner : 4 WUs on a GTX 480


                                                                                                                                                                      Home · Your account · Message boards

                                                                                                                                                                      This material is based upon work supported by the National Science Foundation (NSF) under Grants PHY-1104902, PHY-1104617 and PHY-1105572 and by the Max Planck Gesellschaft (MPG). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the investigators and do not necessarily reflect the views of the NSF or the MPG.

                                                                                                                                                                      Copyright © 2014 Bruce Allen