wonder why?


Advanced search

Message boards : Cruncher's Corner : wonder why?

AuthorMessage
Miklos M.
Send message
Joined: Apr 3 05
Posts: 16
Credit: 13,531,169
RAC: 4
Message 117326 - Posted 19 May 2012 0:06:33 UTC

    I just started up my new computer Windows 7, 3.6 and 8 core versus the older one which is also Windows 7, but 3.2, and 6 core. My Einstein units take longer on the more powerful computer by about 40 minutes each. I have a GTX 580 and the cuda units crunch much faster than on the older one with a GTX 460. That I understand, but I am puzzled by the cores running slower. Does anyone have an answer?

    5pot
    Send message
    Joined: Apr 8 12
    Posts: 109
    Credit: 4,570,419
    RAC: 0
    Message 117328 - Posted 19 May 2012 0:31:37 UTC

      Last modified: 19 May 2012 0:33:18 UTC

      Assuming you have bulldozer. Don't know about this project, but if they do use floating path math, you bulldozer shares those resources within the 4 modules, so each "core" gets 1/2 of a modules FPU (floating point unit).

      Your Phenom 2 however had 6 independent cores, each with its own FPU.

      As a side note, if you're running on all 8 cores+GPU, you should tell BOINC to only use 87.5% of your processors. This will give you GPU more CPU time when it needs it. Should speed others up as well.

      Cheers

      Miklos M.
      Send message
      Joined: Apr 3 05
      Posts: 16
      Credit: 13,531,169
      RAC: 4
      Message 117330 - Posted 19 May 2012 1:29:24 UTC - in response to Message 117328.

        Sorry, but could you repeat it in simpler terms for the novice here, please?

        5pot
        Send message
        Joined: Apr 8 12
        Posts: 109
        Credit: 4,570,419
        RAC: 0
        Message 117331 - Posted 19 May 2012 2:21:25 UTC

          NOT TECHNICAL AT ALL AT THIS STUFF
          I'll try. Not the most technical myself. To put it as simple as I can, floating point can handle, let's say more decimals for the sake of argument than integer math can. integer math may be like 1+1=2 whereas floating point can do 100/9. FOR EXAMPLE (may be wrong it my description, but I'm right below about bulldozer design)

          FPU is can handle more numbers than integer I THINK.

          A bulldozer is made of 8 cores in 4 modules. Each module contains 2 cores. Now each bulldozer core contains its own integer (resources), so if this project used integer math, you would be running 8 integer cores (as far as your CPUs resources go).

          In bulldozer, there is only 1 FPU PER MODULE. Not per core. So, each core must share the FPU resources of the module.

          With your Phenom 6 core, each core had its own FPU. It was not having to share its resources with another core. Thereby, making the tasks run faster.

          Not an engineer (obviously lol), so this is the best I can do. Hope this helps.

          Miklos M.
          Send message
          Joined: Apr 3 05
          Posts: 16
          Credit: 13,531,169
          RAC: 4
          Message 117338 - Posted 19 May 2012 12:30:10 UTC - in response to Message 117331.

            Thank you Spot, it does help and it also made me realize that I should have taken more science courses in college.
            So, in a nutshell, these 8 cores run slower than the 6 cores in the other machine, even though it has a higher Gig processor.
            I have heard that you can make a computer run faster by overclocking it. Is that a safe thing to do?

            Thank you.

            5pot
            Send message
            Joined: Apr 8 12
            Posts: 109
            Credit: 4,570,419
            RAC: 0
            Message 117341 - Posted 19 May 2012 14:45:01 UTC

              Last modified: 19 May 2012 14:46:54 UTC

              BE CAREFUL IF YOU WISH TO DO THIS, YOU HAVE BEEN WARNED

              As long as you have aftermarket cooling than yes. If you're using your stock cooler, it depends.

              Bulldozers tend to be a pretty warm chip, so if you wish to check your temps, I would recommend downloading Core Temp first. See what your temps are. overclock.net says max AMD temps for bulldozer are around 55-60C, so you want to be around 45-50C, since we use these 24/7/365.

              OVERCLOCKING IS SERIOUS BUSINESS TAKE TIME TO LEARN AS MUCH AS YOU CAN, AND GO SLOWLY!!!!!

              You should be able to bring it to 4.0GHz no problem, as long as you have decent cooling.

              Here is a rather basic guide which I think could suit your needs http://apcmag.com/building-a-pc-amd-bulldozer-overclocking-guide.htm

              Don't worry about stress testing, as this is what grid computing (BOINC) is for. With max voltage being 1.4 (vCore), you'll want to stay as under this as possible. The higher the voltage, the shorter the lifespan of the chip. Same applies for temps. The lower the better.

              IF YOU TAKE THE TIME TO LEARN HOW TO OC PROPERLY, YOU WILL LEARN A VERY VALUABLE NEW SKILL. SEARCH INTERNET FORUMS, AND TAKE YOUR TIME. YOU WILL BE SURPRISED AT HOW MUCH YOU CAN LEARN.

              AGAIN, OC AT YOUR OWN RISK

              You will need to pay attention to the multiplier, vcore, and temperature. These will be the most important aspects of overclocking your CPU

              Horacio
              Send message
              Joined: Oct 3 11
              Posts: 205
              Credit: 79,233,790
              RAC: 23,632
              Message 117344 - Posted 19 May 2012 17:47:23 UTC - in response to Message 117338.

                Thank you Spot, it does help and it also made me realize that I should have taken more science courses in college.
                So, in a nutshell, these 8 cores run slower than the 6 cores in the other machine, even though it has a higher Gig processor.
                I have heard that you can make a computer run faster by overclocking it. Is that a safe thing to do?

                Thank you.


                Besides what 5pot said, and that all the overclocking stuff could be an interesting learning, the truth is that your gaining in terms of RAC will not be impressive.
                I used "RAC" not only thinking in the credits itself, but also as a meassure of your contribution to the science of the projects...

                One thing to take in account is that while your phenom take less time to do the work it does at most 6 Wus in that time, while the new CPU will do 8, that's a 33% more WUs, so if the crunching times are not longer than a 33% your will be doing more work anyway...

                ____________

                JLConawayII
                Send message
                Joined: May 12 10
                Posts: 9
                Credit: 11,378,699
                RAC: 0
                Message 117397 - Posted 20 May 2012 22:36:44 UTC

                  Last modified: 20 May 2012 22:37:55 UTC

                  Think of Bull(crap)dozer "8 core" as being a 4 core with poorly designed hardware level hyperthreading and the most bizarrely useless cache design imaginable.

                  5pot
                  Send message
                  Joined: Apr 8 12
                  Posts: 109
                  Credit: 4,570,419
                  RAC: 0
                  Message 117398 - Posted 20 May 2012 23:14:54 UTC

                    Foul- unnecessary roughness.

                    Profile Mike Hewson
                    Forum moderator
                    Avatar
                    Send message
                    Joined: Dec 1 05
                    Posts: 3576
                    Credit: 28,422,730
                    RAC: 8,092
                    Message 117399 - Posted 20 May 2012 23:37:49 UTC

                      Last modified: 21 May 2012 4:33:55 UTC

                      Assuming that aspects of this discussion refer to hyperthreading then see this thread for an interesting discussion. ( Alas I hadn't gotten around to any further Ix as I had alluded to there ). Also beware any effect ( thinking of GW analysis especially ) of variation between the work units themselves - they may not all be created 'equal'.

                      @Miklos : hyperthreading ( in simple terms ) refers to a facility whereby one physical core is 'two-faced' or represents two virtual cores. Think of it like a bank teller that serves two queues. Note that this is a different level concept than 'threads' or 'processes' ( which are operating system abstractions based on varieties of 'task switching' ) as virtual cores have a significant underlying hardware implementation. There is still only one physical core per two virtual cores. All these concepts - cores, threads, processes - refer to degrees of implementation/switching/scheduling of instruction queues ( the lists of instructions that the CPU should perform ) and associated data areas.

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

                      ML1
                      Send message
                      Joined: Feb 20 05
                      Posts: 320
                      Credit: 18,308,720
                      RAC: 9,531
                      Message 117650 - Posted 31 May 2012 0:46:33 UTC - in response to Message 117397.

                        Think of Bull(crap)dozer "8 core" as being a 4 core with poorly designed hardware level hyperthreading and the most bizarrely useless cache design imaginable.

                        Deary me, you obviously have no idea of processor design!

                        Is "hyperthreading" 'crap' because it fools you into believing you have 2 CPUs and yet only delivers 1 CPU and 0.3 of a further CPU?...


                        The Bulldozer design is rather clever and nicely high performance. The limiting factor is the utterly CRAP software and the uneducated Marketing steered unrealistic user expectations.

                        Happy?

                        Now please look again at what can be done and at what price.


                        If you are wanting ultimate performance, then your best bet is to go for an "uber" GPU, and then whatever CPU is adequate to merely feed that GPU.

                        On that count, my research suggests that AMD CPU + nVidia GPU are best for whatever price point you might want.


                        Otherwise, please keep with the original goals of Boinc and acquire whatever computer best fits your needs and then donate only spare CPU/GPU cycles to the furtherance of science.


                        Happy cool crunchin',
                        Martin

                        ____________
                        Powered by Mandriva Linux A user friendly OS!
                        See the Boinc HELP Wiki

                        Miklos M.
                        Send message
                        Joined: Apr 3 05
                        Posts: 16
                        Credit: 13,531,169
                        RAC: 4
                        Message 117716 - Posted 2 Jun 2012 16:33:54 UTC

                          Thank you everyone for trying to explain modern technology to me. I appreciate your taking the time to do so.
                          I think Confucious got his name trying to understand computers, lol.

                          Post to thread

                          Message boards : Cruncher's Corner : wonder why?


                          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