A new Einstein@home CUDA App for Windows is available for Beta Test at Beta Test Page.

We stumbled over some bugs in the CUDA part that might have caused some segfaults, so this is mainly a bugfix release. Also too the CPU part of the App now uses SSE, like in the .09 Beta Apps.

Please test and report, and please include important information (like the NVIdia Driver and Core Client version) in your posts.

BM

Hello all readers,

The prvious version resulting in errors after 2 seconds of running.
This newer version runs ok for now.

Running on i7, 9200 @2.67GHz, 12Gb RAM
GTX285, CUDA version 2.2_vista_64_185.85_general
BOINC version 6.4.7

However I have one question.
BOINC manager has downloaded 4 einsteinbinary_ABP13.07(cuda) jobs, but only one is running at the time. So 6 cores of the pc are doing nothing.
How can I get S5 WU's to crunch?

Thanks.
____________
Greetings from
TJ

How can I get S5 WU's to crunch?

How can I get S5 WU's to crunch?

It could be your BOINC version. You could try a 6.6.x version...

First task back - 136845395 - in just a shade over 6 hours - very similar to previous run, on the same host as I used last time. So no discernable benefit from the SSE code.

Q6600, 9800GT, Windows XP32, drivers 190.38, BOINC v6.6.37

Hi,
I have bought a nvidia 9800 gx2. I am running the 3.10 app. On one gpu it runs perfect but the othe gpu gives immediately a calculation error.
The other thing is, why does the app dont run on the other cpu cores? I´ve got a Q6600 and I am only crunching on 1 cpu +1 gpu. Is it possible to cruch on 2(gpu+cpu) and 2 cpu?

Thanks
hotze
____________

Is it possible to cruch on 2(gpu+cpu) and 2 cpu?

Installed this on my host running Windows 7 64-bit, it started ok!
Will report later when my first result has completed.

System specs as shown by boinc (v. 6.6.36):
20/08/2009 11:29:05 Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU Q9450 @ 2.66GHz [Intel64 Family 6 Model 23 Stepping 7]
20/08/2009 11:29:05 Processor features: fpu tsc pae nx sse sse2 pni mmx
20/08/2009 11:29:05 OS: Microsoft Windows 7: x64 Edition, (06.01.7100.00)
20/08/2009 11:29:05 Memory: 4.00 GB physical, 7.99 GB virtual
20/08/2009 11:29:05 Disk: 50.00 GB total, 16.90 GB free
20/08/2009 11:29:05 Local time is UTC +2 hours
20/08/2009 11:29:06 CUDA device: GeForce 9600 GT (driver version 19038, compute capability 1.1, 512MB, est. 40GFLOPS)

/Holmis
____________

Is it possible to cruch on 2(gpu+cpu) and 2 cpu?

To use the remaining two cores, you'll have to enable a non-GPU application, either from Einstein (includes merging app_info.xml files - for advanced users only) or from another project.

Gruß,
Gundolf

20/08/2009 11:58:41 Einstein@Home Message from server: To get more Einstein@Home work, finish current work, stop BOINC, remove app_info.xml file, and restart.

I get this message and boinc will only ask for work again in 4 hours, but I still got some gpu work which uses a full cpu core too!

Are the gpus running cpu units? what is the speedup of using 1core+1gpu compared to using only 1 core?

Hi,
I have bought a nvidia 9800 gx2. I am running the 3.10 app. On one gpu it runs perfect but the othe gpu gives immediately a calculation error.
The other thing is, why does the app dont run on the other cpu cores? I´ve got a Q6600 and I am only crunching on 1 cpu +1 gpu. Is it possible to cruch on 2(gpu+cpu) and 2 cpu?

Thanks
hotze

20/08/2009 11:58:41 Einstein@Home Message from server: To get more Einstein@Home work, finish current work, stop BOINC, remove app_info.xml file, and restart.

I get this message and boinc will only ask for work again in 4 hours, but I still got some gpu work which uses a full cpu core too!

Are the gpus running cpu units? what is the speedup of using 1core+1gpu compared to using only 1 core?

BTW: app_info.xml of last beta version allowed CPU work, did that change?

My 2 wu with 3.10 were almost exactly the same run time as with previous version.
____________

I can now process GPU work! With the previous 3.09 version, I errored out immediately.

My GPU is NVIDIA GeForce GTX 260 (896MB) driver: 190.38
____________
Regards,
Bob P.

I can now process GPU work! With the previous 3.09 version, I errored out immediately.

My GPU is NVIDIA GeForce GTX 260 (896MB) driver: 190.38

..i have one with 190.38 and will test this version to see if it runs... just got to dl all this stuff.

@RB, what manufacturer is the card?

EDIT!: GTX 260 running app now,
WUIP: http://einstein.phys.uwm.edu/result.php?resultid=136872957

First unit completed under win 7 64-bit on a GeForce 9600GT.
Link to workunit

Did save some time compared to the beta 3.09 CPU-version.
~22000 sec down to ~19300 sec, times are from the web as I'm at work and the host is in my home.
____________

Finished. Win 7 x64, 190.38, 6.6.37.

Is there "pure" CUDA-application is in developer's plans?

Finished. Win 7 x64, 190.38, 6.6.37.

Is there "pure" CUDA-application is in developer's plans?

Ok,
now I´m getting also work for the two remaining cpu cores.
@tjreuter: The 9800GX2 has two chips (2x8800GTS if I´m correct). It´s SLI on one card.
To get the card working at two workunits at a time, I had to disable SLI. Otherwise the second core killed the workunits.
First completed workunits (with 190.38 driver)
57287506
57287483

and first valid
57206708,
which was quite fast. ~14000s for 250 credits. Usually it takes 19000s for the 250 credits units (on a Q6600@3.6GHz).

I like that cuda stuff ;)

hotze
____________

Using CUDA device #0 "GeForce GTX 260" (806.98 GFLOPS)

WT* is that. no offense, but I'm very hesitate to believe those numbers...

holmis's reports 300 gflops as well, is the app mis-reading the gflop number?????????????????????

Ok,
now I´m getting also work for the two remaining cpu cores.
@tjreuter: The 9800GX2 has two chips (2x8800GTS if I´m correct). It´s SLI on one card.

I like that cuda stuff ;)

hotze

Is there "pure" CUDA-application is in developer's plans?

Using CUDA device #0 "GeForce GTX 260" (806.98 GFLOPS)

WT* is that. no offense, but I'm very hesitate to believe those numbers...
what are the specs of that card?????
MHZ?
the damn card better be; Liquid cooled, running at 800 mhz....
please post the message logs of the boinc start-up...
ok, nvm; i'll admit i'm quick to jump to conclusions....
holmis's reports 300 gflops as well, is the app mis-reading the gflop number?????????????????????

Is there "pure" CUDA-application is in developer's plans?

Yes. However, some part of the app will still run on the CPU - it'll never be "pure". We aim to reduce the amount of work done on the CPU as much as we can of course.


Cheers,
Oliver

Is there "pure" CUDA-application is in developer's plans?

Yes. However, some part of the app will still run on the CPU - it'll never be "pure". We aim to reduce the amount of work done on the CPU as much as we can of course.

Using CUDA device #0 "GeForce GTX 260" (806.98 GFLOPS)

WT* is that. no offense, but I'm very hesitate to believe those numbers...
what are the specs of that card?????
MHZ?
the damn card better be; Liquid cooled, running at 800 mhz....
please post the message logs of the boinc start-up...
ok, nvm; i'll admit i'm quick to jump to conclusions....
holmis's reports 300 gflops as well, is the app mis-reading the gflop number?????????????????????

No, it's not. You need to know that this value is computed by the device properties and not determined by benchmarks. It's the card's theoretical maximum performance which can be computed like this: number of multiprocessors * number of cores/multiprocessor * clock rate * flop/core/cycle. For your GTX 260 that's 24 * 8 * 1.4 GHz * 3. By the way, is this an overclocked card? Usually they "only" run at 1.3 GHz...

No one claimed that either your card or any given application could realistically achieve that performance ;-) As always, you may relatively compare different cards using the same measure, despite the absolute values not being very informative.


Cheers,
Oliver

Card is a 9800GT, stock speed, rated at 60 BOINC GFlops, but reported here as 508.03 NVidia marketing GFlops (as also shown in the GPUGrid FAQ). I've commented before that there's a factor of 8 or 9 between 'marketing' and 'crunching' GFlops - it doesn't matter which you use for card comparison purposes, so long as you always state which units you're using, and don't mix them.

new driver available, 190.62, which is working fine on my vista machine.

a new wu run under 190.62. works fine.

Card is a 9800GT, stock speed, rated at 60 BOINC GFlops, but reported here as 508.03 NVidia marketing GFlops (as also shown in the GPUGrid FAQ). I've commented before that there's a factor of 8 or 9 between 'marketing' and 'crunching' GFlops - it doesn't matter which you use for card comparison purposes, so long as you always state which units you're using, and don't mix them.

Ah, so that's why. So 1 BGFlop ~= 8 NVMGFlops

[aside]
Reminiscent of :

- the 1 MB vs 1 million bytes issue ( and GB .... )
- the Cyrix processor's 'Performance Rated' speed ( 'as good as a Pentium' vs actual )
- the quoting of CD drive speeds ( angular or linear velocity? )
[/aside]

Cheers, Mike.

my last cuda wu finished in 5-6 hrs rt.

http://einstein.phys.uwm.edu/result.php?resultid=137117612

my last cuda wu finished in 5-6 hrs rt.

http://einstein.phys.uwm.edu/result.php?resultid=137117612

my last cuda wu finished in 5-6 hrs rt.

http://einstein.phys.uwm.edu/result.php?resultid=137117612

Hi,
is your Q6600 overclocked or running at default speed? I´m looking for some performance numbers between different cards. I want to know, which card gives the most bang for the buck.

hotze

Reminiscent of :

- the 1 MB vs 1 million bytes issue ( and GB .... )

295 is 2*240 core so it's the most bang for buck; but, be sure your power supply can handle it.....

I have crunching a CUDA WU with a GTX 285 in 5,5 hours. My wingmen has crunched the same WU with a CPU in 6 hours.

This is not a big difference .......
____________

@ Einstein@home:

Have you use the latest CUDA to compile, there are some corrections in 2.3
Distributed.net have done this also, bug-fixes and some WU-corruption.
Different field but still a good advise.

I have crunching a CUDA WU with a GTX 285 in 5,5 hours. My wingmen has crunched the same WU with a CPU in 6 hours.

This is not a big difference .......

@ Einstein@home:

Have you use the latest CUDA to compile, there are some corrections in 2.3
Distributed.net have done this also, bug-fixes and some WU-corruption.
Different field but still a good advise.

I have crunching a CUDA WU with a GTX 285 in 5,5 hours. My wingmen has crunched the same WU with a CPU in 6 hours.

This is not a big difference .......

Post the system specs, It's all BETA.
So the more details the better.

Speed should be ecpected (grow in future) better or am i wrong.

If I understand correctly then the current run is a cuda assisted cpu run. The gpu is used for some calculations but not all. Right now your are mainly limited by the cpu. The gpu gives you just a boost of ca. 20-50%.
If your Q9300 is not overclocked then every 4GHz Core2 runs faster than your cpu with cuda help.
When there is a standalone app for the gpu then yours will fly ;)





____________

If I understand correctly then the current run is a cuda assisted cpu run. The gpu is used for some calculations but not all. Right now your are mainly limited by the cpu. The gpu gives you just a boost of ca. 20-50%.
If your Q9300 is not overclocked then every 4GHz Core2 runs faster than your cpu with cuda help.
When there is a standalone app for the gpu then yours will fly ;)

I have been getting the same sort of errors within seconds of starting execution, both on 3.07 and 3.10:

Since we had a new binary for CUDA processing, I thought I'd run a few WUs to see if the situation had improved. Also running 3.05 WUs on the second cpu for the system.

Results:
Currently processing 1 CUDA WU and one cpu only WU. Est. time to completion for cpu WU is 7hrs, 1 min, 9 seconds (they're both almost complete). Est. time to completion for CUDA WU is 7hrs, 53 seconds.

Two previous CUDA WU were logged as taking 25,277.44/25,388.73 secs each.
Three 3.05 WUs that completed also took about the same time to process.

My opinion: the CUDA app (on an AMD X2 5200 system) is not viable presently because it takes both a cpu AND the gpu to process without any noticable improvement in throughput.

AND...if processing SETI on the same system, SETI will monopolize the gpu and not allow E@H to run.

If I understand correctly then the current run is a cuda assisted cpu run. The gpu is used for some calculations but not all. Right now your are mainly limited by the cpu. The gpu gives you just a boost of ca. 20-50%.
If your Q9300 is not overclocked then every 4GHz Core2 runs faster than your cpu with cuda help.
When there is a standalone app for the gpu then yours will fly ;)

I have crunching a CUDA WU with a GTX 285 in 5,5 hours. My wingmen has crunched the same WU with a CPU in 6 hours.

This is not a big difference .......

I have crunching a CUDA WU with a GTX 285 in 5,5 hours. My wingmen has crunched the same WU with a CPU in 6 hours.

This is not a big difference .......

Hello,

I run the einstein cuda-app on a i7 with a GTX285.
It takes about 4.7-5 hours to complete, while I have seen wingman with faster CPU's (non-cuda) and they take 7-8.3 hours to complete. So I see a difference.

Since we had a new binary for CUDA processing, I thought I'd run a few WUs to see if the situation had improved. Also running 3.05 WUs on the second cpu for the system.

Results:
Currently processing 1 CUDA WU and one cpu only WU. Est. time to completion for cpu WU is 7hrs, 1 min, 9 seconds (they're both almost complete). Est. time to completion for CUDA WU is 7hrs, 53 seconds.

Two previous CUDA WU were logged as taking 25,277.44/25,388.73 secs each.
Three 3.05 WUs that completed also took about the same time to process.

My opinion: the CUDA app (on an AMD X2 5200 system) is not viable presently because it takes both a cpu AND the gpu to process without any noticable improvement in throughput.

AND...if processing SETI on the same system, SETI will monopolize the gpu and not allow E@H to run.

I crunched a 3.10 using my CPU only, and found that compared to using the GPU it was about 80% as fast. In other words, the GPU crunches 20% faster than CPU by itself.

As others have commented, the GPU has a lot of potential I believe to become much faster while using less of the CPU. :)

Hi,

i downloaded all as posteted, installed it and started crunching.
Threre are two things i want to report:
fist, the usage of the gpu is low, it reaches only 62 deg (cent.), other applications (gpugrid, if it does not hang, or Seti) heat it to 74 deg.
Three times it caused a complete system hang just when finishing the wu at the beginning of the upload.

Vista64, E8400, 6GB, GTX260 and a lot of troubles.

Regards,

Alexander
____________

To the developers,

The previous days the cuda-app. ran perfect and takes about 4 hours to complete on my i7 with a GTX285. The temperature of the GPU is very low and I like that. Good job.

Today BOINC has downloaded 6 cuda-wu’s, however one can be run at a time. So 7 of the CPU’s are doing now crunching for Einstein and that I do not like. It should be fine when they run 4 to 6 non cuda ABP and/or S5R5 WU’s. Can that be programmed?

Thanks and success.

____________
Greetings from
TJ

This app has started fine on my 9800 GT. Thanks for developing an app compatible with 64-bit Vista/Win7! Other specs: Q9550, BOINC 6.6.36, Driver 190.38, v2.3 dll's.

Runtime projection is ~5 hours compared to 5h 45min with the 3.09 CPU app. So it's in effect slower since the GPU resource is used. First task 137298125.

Following a work request for both CPU and CUDA work I got both but also the 4-hour deferral. The scheduler log is not logical:

2009-08-24 17:28:20.6143 [PID=24659] [HOST#1758399] Sending [RESULT#137355097 h1_0892.30_S5R4__928_S5R5a_1] (est. dur. 14987.39 seconds)
2009-08-24 17:28:20.6177 [PID=24659] [locality] in_send_results_for_file(h1_0892.30_S5R4, 1) prev_result.id=137355097
2009-08-24 17:28:20.6290 [PID=24659] [send] Didn't find anonymous platform app for einstein_S5R5

I've found how to run 4 S5R5 tasks while working on ABP1 on CUDA device.
It's simple. Just freeze all the ABP tasks, then restart BOINC. And after that unfreeze one of ABP tasks and have fun ;)
It seems to run smoothly (but before that I killed some WU's with old driver). Will keep eye on it and report after a while.

Hi Wedge009,

I have been getting the same sort of errors within seconds of starting execution, both on 3.07 and 3.10:

[10:31:00][9612][ERROR] Error creating CUDA FFT plan (error code: 2)
[10:31:00][9612][ERROR] Demodulation failed (error: 3)!
called boinc_finish

Am I missing something really obvious? The host runs SETI CUDA WUs okay. I am not complaining, just reporting - I know improvements can't be made without reports from testers.

I've found how to run 4 S5R5 tasks while working on ABP1 on CUDA device.
It's simple. Just freeze all the ABP tasks, then restart BOINC. And after that unfreeze one of ABP tasks and have fun ;)
It seems to run smoothly (but before that I killed some WU's with old driver). Will keep eye on it and report after a while.

Sorry, not enough memory.

Our final release (scheduler instead of app_info.xml) will contain proper checks that no WU is shipped to you unless your system meets all hardware requirements. FYI, this error is typically caused by on-board devices and/or when your device's global memory is still loaded with (e.g.) textures left by previous 3D apps...

I just wish BOINC wouldn't try to keep asking for GPU work on Einstein, though. The server keeps telling me to remove the app_info.xml, but I'm using the 3.09 ABP beta application since it appears to be show a speed improvement over the current 3.08 release, and it's taking a long time for BOINC to ask for CPU work instead (since there's a four-hour delay after each GPU work request).

Ugh, really? I don't like keeping huge buffers. Unlike some other people, I don't like hogging WUs by having a 10 day cache. ):

Well, I'll keep it in mind for next time, then.
____________
Soli Deo Gloria

Ugh, really? I don't like keeping huge buffers. Unlike some other people, I don't like hogging WUs by having a 10 day cache. ):

Well, I'll keep it in mind for next time, then.

I rarely visit some machines and have to set cache to 10 days and more in case of server failure. Of cause I don't use unstable machines for this.
But for fast machines near me I use a smaller cache - it is easier to change something or even reinstall BOINC if needed without loss of data.

Hi Wedge009,

I have been getting the same sort of errors within seconds of starting execution, both on 3.07 and 3.10:

[10:31:00][9612][ERROR] Error creating CUDA FFT plan (error code: 2)
[10:31:00][9612][ERROR] Demodulation failed (error: 3)!
called boinc_finish

Am I missing something really obvious? The host runs SETI CUDA WUs okay. I am not complaining, just reporting - I know improvements can't be made without reports from testers.

Sorry, not enough memory.

Our final release (scheduler instead of app_info.xml) will contain proper checks that no WU is shipped to you unless your system meets all hardware requirements. FYI, this error is typically caused by on-board devices and/or when your device's global memory is still loaded with (e.g.) textures left by previous 3D apps...

Cheers,
Oliver

Hi Wedge009,

I have been getting the same sort of errors within seconds of starting execution, both on 3.07 and 3.10:

[10:31:00][9612][ERROR] Error creating CUDA FFT plan (error code: 2)
[10:31:00][9612][ERROR] Demodulation failed (error: 3)!
called boinc_finish

Am I missing something really obvious? The host runs SETI CUDA WUs okay. I am not complaining, just reporting - I know improvements can't be made without reports from testers.

Sorry, not enough memory.

Our final release (scheduler instead of app_info.xml) will contain proper checks that no WU is shipped to you unless your system meets all hardware requirements. FYI, this error is typically caused by on-board devices and/or when your device's global memory is still loaded with (e.g.) textures left by previous 3D apps...

Cheers,
Oliver

Hi Oliver

I wonder if you someone would be kind enough to state what the system requirements are which are planned to go into the scheduler. I just wasted a few hours getting setup to "Play" with the Beta on my 8400GS to get the same error on 2 WU's

Regards

My first WU just completed in 9 hours with no problems.
The power draw of my PC whilst crunching this unit is around 160 Watts, as opposed to around 200 Watts when crunching CUDA SETI or GPUGRID units.

System info:
Windows Vista Home Premium SP1 64-bit
BOINC version 6.6.36
NVIDIA driver version 190.38
GeForce GTX 280
AMD Athlon 64 3200+
4GB RAM
____________

Well, I put 10 days for connection and 10 days for work cache... and still Einstein is only asking work for the GPU and waiting 4 hours thereafter. This is... a little frustrating.

Edit: Never mind, BOINC finally asked for Einstein CPU WUs... but only after it managed to scrounge one SETI GPU WU. For the record, this is BOINC 6.6.36, the latest stable release.
____________
Soli Deo Gloria

...For the record, this is BOINC 6.6.36, the latest stable release.

I've given up. To get Einstein CPU WUs, I have to leave the GPU information in the app_info.xml and just put up with the repeated errors on the GPU WUs. As long as there are no GPU WUs across all projects, the scheduler will always ask for GPU WUs first. ):
____________
Soli Deo Gloria

I've given up. To get Einstein CPU WUs, I have to leave the GPU information in the app_info.xml and just put up with the repeated errors on the GPU WUs. As long as there are no GPU WUs across all projects, the scheduler will always ask for GPU WUs first. ):

You misunderstand. That's what I've been trying to do. 3.09 works great and seems to be a substantial improvement over 3.08, so that's why I've been trying to run it instead of just removing the app_info.xml altogether as the Einstein server keeps yelling at me to do. But no matter what I try, BOINC insists on giving priority to requesting GPU WUs before CPU ones. I strongly recommend releasing 3.09, it seems to be thoroughly tested now.
____________
Soli Deo Gloria

You misunderstand. That's what I've been trying to do ....

• Lots of people on many projects are complaining about scheduling problems in 6.6.36.
  
• If you don't want GPU capability, something like 6.4.7 or 6.5.0 should be much better for you.
  
• Setting 10/10 for your cache sizes is likely to worsen the already bad behaviour of 6.6.36. If you are 'always on' use something like 0/1 to 0/4.
  
• The EAH project has some sort of server bug that causes the scheduler to say randomly that it has no tasks available even if it has just given you some. With S5R5 this causes a 1 min backoff before a retry is done. With ABP1, the backoff is 4 hours. You don't have to endure the 4 hour backoff. Just wait 1 min and then force a retry with the 'update' button. Once you have sufficient tasks on board, the backoff shouldn't bother you any further.
  

If you need help with transitioning to the CPU only app_info.xml just ask. If you are not careful you can trash your current cache. If you have no EAH GPU tasks on board, you should be able to stop BOINC, uninstall BOINC, copy the CPU app_info.xml and the CPU ABP1 binaries over the top of the existing ones, overwriting existing files if the names are the same and then install a more suitable version of BOINC, as suggested above, to replace 6.6.36. You can get any particular version of BOINC you fancy from here.

____________
Cheers,
Gary.

First of all, I meant no offence. I know this is all volunteer-run.

Maybe I'm missing something but if you really are trying to get CPU tasks rather than GPU tasks, why don't you ditch the GPU app_info.xml and set up the CPU one instead?

You said earlier that you "have to leave the GPU information in the app_info.xml". I'm sorry but this is quite wrong - you certainly don't have to do that.

Lots of people on many projects are complaining about scheduling problems in 6.6.36.

If you don't want GPU capability, something like 6.4.7 or 6.5.0 should be much better for you.

Setting 10/10 for your cache sizes is likely to worsen the already bad behaviour of 6.6.36. If you are 'always on' use something like 0/1 to 0/4.

The EAH project has some sort of server bug that causes the scheduler to say randomly that it has no tasks available even if it has just given you some. With S5R5 this causes a 1 min backoff before a retry is done. With ABP1, the backoff is 4 hours. You don't have to endure the 4 hour backoff. Just wait 1 min and then force a retry with the 'update' button. Once you have sufficient tasks on board, the backoff shouldn't bother you any further.

[quote]* If you don't want GPU capability, something like 6.4.7 or 6.5.0 should be much better for you.

And how is the the 3.10 doing.

Any better then the 80%. in 3.09

As in 100% with GPU the CPU would be @ 80%.

First of all, I meant no offence. I know this is all volunteer-run.

Maybe I'm missing something ...

Because I had already set up a CPU-only app_info.xml previously....

<sigh> In any case, hopefully my woes will all become moot shortly. I'm waiting on a replacement video card to come in which will surely be sufficient for processing the Einstein GPU WUs.

It appears that BOINC detects the presence of a GPU on this particular host and will request GPU units regardless of the fact that there is no GPU application included in the app_info.xml. And it is because of the fact that there are no GPU applications in the app_info.xml that the Einstein server responds with the "remove app_info.xml" message and will then set a four-hour delay for the next communication with it.

I seemed to get a 4-hour wait every time, not the increasing intervals each time as you described.

It's okay now, anyway. Got my new GPU installed and working happily now. Just would have been a concern if I remained with the weaker GPU, though...
____________
Soli Deo Gloria

Tested with Nvidia driver version 190.62 on GeForce GTX 280.
BOINC CC 6.6.36

WU completes successfully with no problems.
____________

Hi,
from time to time I get the following error:
<core_client_version>6.10.4</core_client_version>
<![CDATA[
<message>
Das System kann den angegebenen Pfad nicht finden. (0x3) - exit code 3 (0x3)
</message>
<stderr_txt>
Activated exception handling...
[08:31:04][6088][INFO ] Starting data processing...
[08:31:04][6088][INFO ] Using CUDA device #0 "GeForce 9800 GX2" (642.82 GFLOPS)
[08:31:04][6088][ERROR] Couldn't open template bank file: templates_400Hz_2.bank (No such file or directory).
[08:31:04][6088][ERROR] Demodulation failed (error: 3)!
called boinc_finish

</stderr_txt>
]]>

I have changed to to new boinc version (from 6.6.36) because I had the same problem.
Earlier I got his problem:
[00:18:22][5104][ERROR] Error creating CUDA FFT plan (error code: 2)
[00:18:22][5104][ERROR] Demodulation failed (error: 3)!
On the host there is only einstein@home working on the gpu. No games or other 3d stuff.
System:
WinXP SP3 32bit, q6600, 9800GX2 with 190.38 driver, SLI disabled
Any suggestions?
____________

Hi,
from time to time I get the following error:
<core_client_version>6.10.4</core_client_version>
<![CDATA[
<message>
Das System kann den angegebenen Pfad nicht finden. (0x3) - exit code 3 (0x3)
</message>
<stderr_txt>
Activated exception handling...
[08:31:04][6088][INFO ] Starting data processing...
[08:31:04][6088][INFO ] Using CUDA device #0 "GeForce 9800 GX2" (642.82 GFLOPS)
[08:31:04][6088][ERROR] Couldn't open template bank file: templates_400Hz_2.bank (No such file or directory).
[08:31:04][6088][ERROR] Demodulation failed (error: 3)!
called boinc_finish

</stderr_txt>
]]>

I have changed to to new boinc version (from 6.6.36) because I had the same problem.
Earlier I got his problem:
[00:18:22][5104][ERROR] Error creating CUDA FFT plan (error code: 2)
[00:18:22][5104][ERROR] Demodulation failed (error: 3)!
On the host there is only einstein@home working on the gpu. No games or other 3d stuff.
System:
WinXP SP3 32bit, q6600, 9800GX2 with 190.38 driver, SLI disabled
Any suggestions?

I am a q9400, win7, 4GB ram, GTX275

The following wu went into error:

140476115
140474992
140193504
140164839
____________

the following wu got an error immediately (tried also updating cudart and cudafft dlls to 2.3 but didn't work)

141150679
141150578
141150530
141150497
141150473
141150232

GeForce 8800 GTS with 320MB (compute capability 1.0, cuda 2.3, driver 190.62)
____________

I just want to share my experience with the cuda app. As I mentioned earlier about half of the workunits get an calculation error:

Exit status 3 (0x3)

<core_client_version>6.10.4</core_client_version>
<![CDATA[
<message>
Das System kann den angegebenen Pfad nicht finden. (0x3) - exit code 3 (0x3)
</message>
<stderr_txt>
Activated exception handling...
[02:13:28][2040][INFO ] Starting data processing...
[02:13:28][2040][INFO ] Using CUDA device #1 "GeForce 9800 GX2" (642.82 GFLOPS)
[02:13:28][2040][INFO ] Checkpoint file unavailable: status.cpt (No such file or directory).
------> Starting from scratch...
[02:13:28][2040][INFO ] Header contents:
------> Original WAPP file: p2030_53647_84633_0048_G55.97-00.86.N_4.wapp
------> Sample time in microseconds: 128
------> Observation time in seconds: 268.9792
------> Time stamp (MJD): 53647.979548611111
------> Number of samples/record: 512
------> Center freq in MHz: 1420
------> Channel band in MHz: 0.390625
------> Number of channels/record: 256
------> Nifs: 1
------> RA (J2000): 193815.465068
------> DEC (J2000): 200556.780691
------> Galactic l: 56.0573
------> Galactic b: -0.7369
------> Name: G55.97-00.86.N
------> Lagformat: 0
------> Sum: 1
------> Level: 3
------> AZ at start: 470.4938
------> ZA at start: 4.9436
------> AST at start: 0
------> LST at start: 0
------> Project ID: p2030
------> Observers: Arzoumanian
------> File size (bytes): 16190702
------> Data size (bytes): 16179201
------> Number of samples: 2097152
------> Trial dispersion measure: 420.4 cm^-3 pc
------> Scale factor: 8060.69
[02:13:29][2040][INFO ] Seed for random number generator is -1132392027.
[02:13:30][2040][ERROR] Error creating CUDA FFT plan (error code: 2)
[02:13:30][2040][ERROR] Demodulation failed (error: 3)!
called boinc_finish

</stderr_txt>

I have updated BOINC to 6.10.4 from 6.6.36 because I had the problem before. I also updated the nvidia driver to the current 190.62 on a 9800GX2 (2x512MB).
The behaviour of BOINC is:
first: killing a number of workunits according the number of gpus. (I have updated to a second 9800GX2 and now it is killing 4 instead of 2 workunits).
second: calculating 4 workunits to 100%, then killing 4 workunits.
now it is in a state where there are 4 wu @ 94%, 4 wu @ 80%, 2 wu @ 4%, 3 wu @ 1 % and 16 wu < 1%. status in every line is suppressed/displaced (german: verdrängt).
Before I had upgraded to two 9800GX2 the situation was the following:
one gpu was crunching at one wu, but the second keep jumping between the other wu according to: starting by a low % wu and whenever the accomplished % equals the % the next wu then it suspended the current wu and jumped to the next one. This keeps happeniing up to a high % wu. Then it marked a new wu with suppressed and started all over again.

For me it seems more like a scheduler problem. The result is running out of memory on the gpu and than have no empty ram and so killing a wu.

Now I will change the OS from WinXP32bit SP3 to Win764bitRC and see, if there is the same.
hotze

____________

...For me it seems more like a scheduler problem. The result is running out of memory on the gpu and than have no empty ram and so killing a wu.

...For me it seems more like a scheduler problem. The result is running out of memory on the gpu and than have no empty ram and so killing a wu.

Yes, and if I recall correctly, it was fixed in 6.6.38 and a 6.10.x version after 6.10.4.

Gruß,
Gundolf

...For me it seems more like a scheduler problem. The result is running out of memory on the gpu and than have no empty ram and so killing a wu.

Yes, and if I recall correctly, it was fixed in 6.6.38 and a 6.10.x version after 6.10.4.

Gruß,
Gundolf

Ok I will give it a try. But 6.6.38 has another bug (atleast in win7 64bit). It only recognizes 1 graphicscard. The second card is not detected. 6.5 is working fine so far.
I will try 6.6.38 in winxp later.

hotze

Not detected, or detected but not used?

If you have two graphics cards of different specifications, BOINC v6.6.xx (and v6.10.xx) is designed only to use the "most capable". You can change that by setting the <use_all_gpus>1</use_all_gpus> configuration flag documented in client configuration.

BOINC 6.6.38 (winxp32 sp3) hasn´t trashed any wu so far. But I have seen a runtime variance between the gpus in the PCIe16 slot and the PCIe4 slot (which is reasonable).
PCIe16: 14200 seconds
PCIe4: 15400 seconds
on a Q6600 @ 3.6GHz.
____________

Just so you folks know, for 6.10.x the only "usable" versions are .3, .7, and .11 based on experiences on GPU Grid, Collatz, MW ... oh, and my own personal use. :)

Though a few people have had .6 work for them there are serious issues with that build for most people. And yes, task deaths are among the problems with several of those versions ... as is unconstrained downloads.

I am mostly running 6.10.11 and 6.10.7 -- I think I have a few other iterations out there -- probably need to address that -- though I tend not to change until properly burned on a specific workstation...

Just so you folks know, for 6.10.x the only "usable" versions are .3, .7, and .11 based on experiences on GPU Grid, Collatz, MW ... oh, and my own personal use. :)

Though a few people have had .6 work for them there are serious issues with that build for most people. And yes, task deaths are among the problems with several of those versions ... as is unconstrained downloads.

Quite all my CUDA Wu are aborting with a "Compute error" after 4-5 elapsed hours.
Now I have 21 WUs in the tube.
I think that if also the next one will abort, I'll force the abort for the other WUs, and I'll wait for a next ABP1 version.

Win XP 64b. BM 6.10.11 64b version. CUDA 190.62. NVIDIA 9800GT
____________

No reason to change if there is no problem or you don't need a feature. I had been moving up because of the need for ATI support and then left it on a couple machines even after I moved cards about.

I have at least two machines on 6.6.2x I forget the exact versions ... for one thing on the Mac I don't know that there is any improvement in moving up ...

Quite all my CUDA Wu are aborting with a "Compute error" after 4-5 elapsed hours...

Quite all my CUDA Wu are aborting with a "Compute error" after 4-5 elapsed hours...

Yew I saw it. But I can't do anything.
Another WU with Compute Error.
I aborted all the remaining WU
____________

"Maximum elapsed time exceeded"

Yes I know about it. But I can't do anything.

In next few days I'll receive a new GTX260. I hope it will better !!
____________

"Maximum elapsed time exceeded"

Yep, I've just had two in a row do the same thing after five hours, don't think my 8800gtx is up to it possibly!

:O(
____________

how much % improvement to CPU only.

seems a bit slow on improved clients.
3.10 was from 19 aug. gettign almost 2 months old ?!!!

and no questions from the developer about bugs that get reported.

Hi!

I forwarded the info on the "resource limit exceeded" problem to the devs.

The ABP1 apps are configured to require at most 1.1 Peta Floating Point Operations .... which is ok if BOINC is using this figure against the CPU benchmarks. If BOINC (maybe only in newer versions) figures these are CUDA floating point ops, then this could be a problem...some cards have 800GFlops and more throughput.

In the meantime until the devs have responded: if you are courageous, you can try the following (experts only, and only if you get this nasty -177 error!!!:

* Stop BOINC
* edit the client_state.xml file :
**** find tags <rsc_fpops_bound>1100000000000000.000000</rsc_fpops_bound>
**** add two more zeros after the 11
* save client_state.xml
* restart BOINC.

Please report if this helped.

Thx
Bikeman
____________

@Bikeman

Ok It works !!!!!!!!!!
____________

@Bikeman

Ok It works !!!!!!!!!!

what is the hidden point of use expensive graphic cards and get boring 20% of perfomance increase?
for example milkyway@home get up to 60x (6000%) !! performance increase by using GPU.

what is the hidden point of use expensive graphic cards and get boring 20% of perfomance increase?
for example milkyway@home get up to 60x (6000%) !! performance increase by using GPU.

It's not "hidden", it's public...

i mean why so small advantage of using gpu?

We will tune the application for maximum performance as we go...
Oliver

...In the meantime until the devs have responded: if you are courageous, you can try the following (experts only, and only if you get this nasty -177 error!!!:

* Stop BOINC
* edit the client_state.xml file :
**** find tags <rsc_fpops_bound>1100000000000000.000000</rsc_fpops_bound>
**** add two more zeros after the 11
* save client_state.xml
* restart BOINC...

The splitters provide both <rsc_fpops_est> used to estimate runtime, and 10 times that as <rsc_fpops_bound> used to determine when work has run so long that BOINC should kill it with a "Maximum elapsed time exceeded". BOINC uses DCF in the estimate but not in the max. If DCF were 1, a VLAR which wasn't rebranded would be killed at 10 times the estimate, if DCF were 0.1 it would have to run 100 times the estimate to reach the max limit.

Hi Gundolf,

Thanks for the (as always) excellent advice, I didn't know about that <flops> thing in the app_info.xml schema.

So, as the cuda beta app's runtime is in the same order of magnitude as the CPU variant, would it make sense to set <flops>x</flops> then to (say) half the value of <p_fpops> in your client_state.xml ?

CU
Bikeman

____________

started my first einstein beta unit on my homeserver with 9600 GT card
running it under 6.10.13 boinc and using the latest version provided

i am gonna see if this is gonna run as good as it does on gpugrid or collatz.

but by reading through the messages i see not much performance gain being reported.
while on most cuda/ati projects enormous gains are being reported when using the cuda/ati cards.
____________

started my first einstein beta unit on my homeserver with 9600 GT card
running it under 6.10.13 boinc and using the latest version provided

i am gonna see if this is gonna run as good as it does on gpugrid or collatz.

but by reading through the messages i see not much performance gain being reported.
while on most cuda/ati projects enormous gains are being reported when using the cuda/ati cards.

...So, as the cuda beta app's runtime is in the same order of magnitude as the CPU variant, would it make sense to set <flops>x</flops> then to (say) half the value of <p_fpops> in your client_state.xml ?

David Anderson checked in changeset [19282] less than an hour ago, apparently in response to this problem.

His solution: assume the CUDA card runs at the same speed as the CPU, unless app_info has a flops line to the contrary. While waiting for a new BOINC client to be compiled, you could test it by putting your CPU floating point benchmark figure in app_info and see how well his idea stacks up!

Make sure you get the orders of magnitude right. The figure to put in app_info is measured in individual FLOPs (billions of them). Benchmarks are usually stated in MFLOPs (thousands of them), and CPU speeds in GFlops (not very many of them at all). You can use exponential format in app_info, I'm told, but Bikeman's suggestion of pulling <p_fpops> out of client_state is probably the easiest and most reliable.

Indeed, that change is related to the discussion here and is basically reverting an earlier change. If you are using a BOINC version older than 6.10.5 you probably are not affected by this problem anayway (problem = result terminated prematurely by BOINC because BOINC thinks it is in an endless loop as it's resource limit was exceeded).

CU
Bikeman
____________

Task ID 142450599
Name p2030_53995_04256_0033_G63.04-02.74.C_5.dm_443_1
Workunit 59849668
Created 7 Oct 2009 8:32:53 UTC
Sent 7 Oct 2009 19:40:50 UTC
Received 8 Oct 2009 3:21:09 UTC
Server state Over
Outcome Client error
Client state Compute error
Exit status -177 (0xffffffffffffff4f)
Computer ID 2086650
Report deadline 21 Oct 2009 19:40:50 UTC
CPU time 0
stderr out <core_client_version>6.10.13</core_client_version>
<![CDATA[
<message>
Maximum elapsed time exceeded
</message>
]]>

Validate state Invalid
Claimed credit 0
Granted credit 0
application version 3.10

____________

Task ID 142450599
Name p2030_53995_04256_0033_G63.04-02.74.C_5.dm_443_1
Workunit 59849668
Created 7 Oct 2009 8:32:53 UTC
Sent 7 Oct 2009 19:40:50 UTC
Received 8 Oct 2009 3:21:09 UTC
Server state Over
Outcome Client error
Client state Compute error
Exit status -177 (0xffffffffffffff4f)
Computer ID 2086650
Report deadline 21 Oct 2009 19:40:50 UTC
CPU time 0
stderr out <core_client_version>6.10.13</core_client_version>
<![CDATA[
<message>
Maximum elapsed time exceeded
</message>
]]>

Validate state Invalid
Claimed credit 0
Granted credit 0
application version 3.10

Hi!

This error seems to be caused by what we were discussing a few messages ago in this thread: Newer BOINC versions will perform the calculation of the maximum runtime per result differently.

To fix this, you can try the following:

- shutdown Boinc
- Find the app_info.xml file, it should be in it should be "C:\Documents and Settings\All Users\Application Data\BOINC\projects\einstein.phys.uwm.edu" or similar.
-insert this line after the line ending in </plan_class>:
<flops>3000000000.0</flops>
- save modified app_info.xml
- restart BOINC

Hope this helps,
Bikeman

I want to crunch einstein only on the gpu and don´t want to download any cpu work. Is this possible?
I´ve been crunching gpu tasks for the last 3 weeks and no cpu one. But the scheduler keeps downloading cpu tasks and reports them as aborted after 2 weeks.
____________

I stopped this beta since every unit gets aborted or reported as in error
The funny thing is these program uses both gpu and cpu and takes more time to complete as a normal unit on the same cpu
Normal unit finished in 6 to 7 hours the so called gpu unit passes 8 hours with ease
____________

I want to crunch einstein only on the gpu and don´t want to download any cpu work. Is this possible?
I´ve been crunching gpu tasks for the last 3 weeks and no cpu one. But the scheduler keeps downloading cpu tasks and reports them as aborted after 2 weeks.

I want to crunch einstein only on the gpu and don´t want to download any cpu work. Is this possible?

Maybe a bit over the top for the developers.

On http://gpgpu.org/
there is a news article about webinars you can follow.
And previous once to view from Nvidia about Cuda and openCL.

Ati is also going openCL with a new SDK.

And S3 launched a low-power S3 Graphics 5400E wich will support openCL

i mean why so small advantage of using gpu?

We will tune the application for maximum performance as we go...

Oliver

how long shall we wait for this application?

Some more speed, not easy oke, will get you more users.
Wich will give more response on the bugs and responsiveness and performance.


An attempt to see how you use the GPU.
You see at the ABP1-program and see what calculations the GPU can take over.
Or lett the GPU calculate all and use the CPU as a co-proccessor.

As the GPU has parallel mass processing and the CPU the handling tools.

aka a program is difficult to copy from 1 architeture to another.
Withouth trowing it over and start from a new point.

how long shall we wait for this application?

The engineer in me is tempted to say "'till it's ready" :-). Seriously tho, remember that the ABP1 Binary Pulsar Search is relatively young: While E@H started several years ago, the ABP1 search began only in March this year. CUDA support in BOINC also is rather new, and for my personal taste, Boinc 6.10.x is the first release that really handles GPU tasks reliably wrt. task scheduling, resource shares etc.

CU
Bikeman

Just finished the first 2 WUs and from the lloks of it, they were kind of slow http://einstein.phys.uwm.edu/results.php?userid=271017
My xeon @ 2.5 GHz was running at 80% speed.
For the 3.10 app, what is the estimated speed-up compared to the CPU version?

Variable, and probably depends significantly on the speed of your CUDA card too.

Your Quadro FX 580 isn't listed in the developer documentation yet (at least not the public release version - see page 110 of that 145 page PDF document) - but its predecessor the Quadro FX 570 is very much at the bottom of the range - it has only 4 multiprocessors, where other consumer-grade cards have up to 30.

Overall, I would expect that the later CPU-only v3.11 Einstein ABP1 release would probably run faster on your machine that this initial CUDA test version. But maybe the balance will shift with the next CUDA release.

BTW, Bernd, how are you getting on? Anything new for us to test in the near future?

11.11.2009 18:50:20 Einstein@Home Sending scheduler request: Requested by user.
11.11.2009 18:50:20 Einstein@Home Requesting new tasks for CPU
11.11.2009 18:50:35 Einstein@Home Scheduler request completed: got 0 new tasks
11.11.2009 18:50:35 Einstein@Home Message from server: To get more Einstein@Home work, finish current work, stop BOINC, remove app_info.xml file, and restart.
11.11.2009 18:50:35 Einstein@Home Message from server: No work sent
11.11.2009 18:50:35 Einstein@Home Message from server: Your app_info.xml file doesn't have a version of Hierarchical S5 all-sky GW search #6.


What's wrong?

11.11.2009 18:50:04 Starting BOINC client version 6.10.18 for windows_x86_64
11.11.2009 18:50:04 Processor: 2 AuthenticAMD AMD Athlon(tm) 64 X2 Dual Core Processor 5200+ [AMD64 Family 15 Model 67 Stepping 2]
11.11.2009 18:50:04 Processor: 1.00 MB cache
11.11.2009 18:50:04 OS: Microsoft Windows 7: Ultimate x64 Edition, (06.01.7106.00)
11.11.2009 18:50:04 NVIDIA GPU 0: GeForce GTX 260 (driver version 19107, CUDA version 2030, compute capability 1.3, 896MB, 497 GFLOPS peak)
11.11.2009 18:50:04 Einstein@Home Found app_info.xml; using anonymous platform

11.11.2009 18:50:20 Einstein@Home Requesting new tasks for CPU
...
11.11.2009 18:50:35 Einstein@Home Message from server: Your app_info.xml file doesn't have a version of Hierarchical S5 all-sky GW search #6.

What's wrong?

2. Does your app_info.xml contain S5R6-related information?

Gruß,
Gundolf

2. Does your app_info.xml contain S5R6-related information?

Gruß,
Gundolf

Don't know. It's from newly downloaded archive with cuda apps (einsteinbinary_ABP1_3.10_windows_intelx86_cuda.zip).

What should be there else?

Thanx, it started work.

But why not to insert these changes into archive at "Einstein@Home Beta Testing" page?

And else, should I change an app_info on other machine, pure-CPU host?

And else, should I change an app_info on other machine, pure-CPU host?

Hello,
I download (automaticaly) the new cuda zip archive today from Einstein.
Using the GPU slew down my computer graphics !
I installed the latest driver for my graphic card (Nvidia - 9400 GP)
intalled the latest Boinc version 6.10.18 and... my graphics still moving VERY slowly.. SoI went back to my last Boinc version 6.6.36... nothing change.

In Boinc preferences, everything return normal when I unchecked use the GPU while the computer is in use.

Can you explain me how to come back to the last-1 version of your cuda driver ?
In advance, thanks a lot.
Lionel.

In fact it takes a few seconds to slow down the graphics. Seems like the memory of the GPU is getting full and the computer spend is time to swap between the GPU memory and the rest... All the graphics a terribly slow (close window, open window...)

Yes, I forgot in my last email : I'm using a Pentium IV 3.06 GHz, 1 G Ram, and Nvidia GPU 9400 GP.
Bests regards.
Lionel.

Hello,
I download (automaticaly) the new cuda zip archive today from Einstein.
Using the GPU slew down my computer graphics !
I installed the latest driver for my graphic card (Nvidia - 9400 GP)
intalled the latest Boinc version 6.10.18 and... my graphics still moving VERY slowly.. SoI went back to my last Boinc version 6.6.36... nothing change.

In Boinc preferences, everything return normal when I unchecked use the GPU while the computer is in use.

Can you explain me how to come back to the last-1 version of your cuda driver ?
In advance, thanks a lot.
Lionel.

Hello Gul,

Thanks for answering me.

I'm using the GPU since the beginning with the Seti project.
It works fine with the 6.6.36 Boinc version.
I joined Einstein several weeks ago, running both Seti and Einstein on my first Computer. It works fine.

I let the computer handle the priority between Seti and Einstein WU and the GPU is only crunching Seti WU. The CPU is crunching one Einstein WU and one Seti WU.

6 november : I changed my Einstein preferences. Nothing changes. I checked "No requesting new task".
12 november, I finished my WU and unchecked "no requesting new task". I saw that Einstein downloaded two big files (around 20 Mb), intel x86 something and graphics something. In the evening - I switch off the computer for two hours -, I saw that the screening started to slow down terribly. I tryed to change my Einstein preferences - unsuccessfull : the Boinc messages show that the changes seems to be ignored -, then I upgrade Boinc, I upgrade my GPU driver... nothing change.

I'm waiting to finish my actual WU then I'll quit Einstein project to subscribe again. I hope the new files downloaded will change the effect.

Bsts rgrds
Lionel (Kty is my wife and I'm using her computer 8-)

Hello,
I download (automaticaly) the new cuda zip archive today from Einstein.
Using the GPU slew down my computer graphics !

Does the situation improve if you temporarily halt (suspend) all running E@H jobs, and does it go back to "slow" if you resume them?

Cheers
Bikeman

I Tried to use the app_info.xml from the beta page but it wouldnt run. I need entries for S5R6 and Arecibo in the file as well.

I dont know what to enter into the file, could someone advise what to do please.

I Tried to use the app_info.xml from the beta page but it wouldnt run. I need entries for S5R6 and Arecibo in the file as well.

I dont know what to enter into the file, could someone advise what to do please.

I Tried to use the app_info.xml from the beta page but it wouldnt run. I need entries for S5R6 and Arecibo in the file as well.

I dont know what to enter into the file, could someone advise what to do please.

Already posted (details in message 99970)