Forum > GPU crunching
V10/11 of modified SETI MB CUDA + opt AP package for full multi-GPU+CPU use
k6xt:
I'm successfully processing only AP and CUDA WU on a pair of Q6600 with CUDA cards using the optimized apps. Over the past few weeks my RAC has doubled to about 16,000 mostly due to these two PCs. All seems to be working fine except:
The stderr files shows CUDA's are taking on the order 100 seconds. But the calendar time seems to be a few hours. The CUDA cards run about as hot as usually stated on this site. I'm hoping an expert can spot why the CUDA's seem to take up excessive calendar time. Appreciate any help!
Here is the stderr for a typical work unit:
http://setiathome.berkeley.edu/result.php?resultid=1208947940
and for the same WU this is the BOINC Message showing almost 2 hours to complete:
2009-04-27 05:00:25 SETI@home Starting task 01fe09aa.6562.18886.6.8.69_0 using setiathome_enhanced version 608
2009-04-27 06:56:12 SETI@home Computation for task 01fe09aa.6562.18886.6.8.69_0 finished
At these times the PC was idle and, in any case, I permit CUDA computation while the PC is in use.
The SAH portion of app_info is:
<app>
<name>setiathome_enhanced</name>
</app>
<file_info>
<name>AK_v8_win_SSSE3x.exe</name>
<executable/>
</file_info>
<file_info>
<name>setiathome_6.08_windows_intelx86__cuda.exe</name>
<executable/>
</file_info>
<file_info>
<name>cudart.dll</name>
<executable/>
</file_info>
<file_info>
<name>cufft.dll</name>
<executable/>
</file_info>
<file_info>
<name>libfftw3f-3-1-1a_upx.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>603</version_num>
<platform>windows_intelx86</platform>
<flops>5160439018.79946125</flops>
<file_ref>
<file_name>AK_v8_win_SSSE3x.exe</file_name>
<main_program/>
</file_ref>
</app_version>
<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>608</version_num>
<platform>windows_intelx86</platform>
<avg_ncpus>0.127970</avg_ncpus>
<max_ncpus>0.127970</max_ncpus>
<flops>1000000000</flops>
<plan_class>cuda</plan_class>
<file_ref>
<file_name>setiathome_6.08_windows_intelx86__cuda.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>cudart.dll</file_name>
</file_ref>
<file_ref>
<file_name>cufft.dll</file_name>
</file_ref>
<file_ref>
<file_name>libfftw3f-3-1-1a_upx.dll</file_name>
</file_ref>
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>
</app_version>
</app_info>
Richard Haselgrove:
--- Quote from: k6xt on 27 Apr 2009, 11:56:44 am ---I'm successfully processing only AP and CUDA WU on a pair of Q6600 with CUDA cards using the optimized apps. Over the past few weeks my RAC has doubled to about 16,000 mostly due to these two PCs. All seems to be working fine except:
The stderr files shows CUDA's are taking on the order 100 seconds. But the calendar time seems to be a few hours. The CUDA cards run about as hot as usually stated on this site. I'm hoping an expert can spot why the CUDA's seem to take up excessive calendar time. Appreciate any help!
Here is the stderr for a typical work unit:
http://setiathome.berkeley.edu/result.php?resultid=1208947940
and for the same WU this is the BOINC Message showing almost 2 hours to complete:
2009-04-27 05:00:25 SETI@home Starting task 01fe09aa.6562.18886.6.8.69_0 using setiathome_enhanced version 608
2009-04-27 06:56:12 SETI@home Computation for task 01fe09aa.6562.18886.6.8.69_0 finished
At these times the PC was idle and, in any case, I permit CUDA computation while the PC is in use.
--- End quote ---
The line in stderr says "CPU time 139.5781"
That's how long the Q6600 spent on them. Your 8500 GT (unfortunately, one of the slowest CUDA cards available) may indeed have taken almost two hours to complete the task. But during that time, all but a minute fraction of the power of one core of the Q6600 was available for AP, browsing the web, or whatever. So the CUDA contribution, comparatively small as it is, is all a net gain to your productivity.
Raistmer:
--- Quote from: aradzh on 26 Apr 2009, 11:03:18 pm ---Is there is a possibility to suspend only GPU computation while use computer with moded CUDA ? With stock client I uncheck "Use GPU while computer is in use" in preferences of client but with moded aplications it is not work :( and only possiblity seems to uncheck "While computer is in use" with small idle, say 0.10 ;)
Thanks in advance,
--- End quote ---
Try to switch to BOINC 6.6.20 + corresponding app_info. Now it's recommended method of using modded SETI CUDA MB. "team" mod considered as deprecated :)
--- Quote ---Andrey
--- End quote ---
По русски читать можешь?
Raistmer:
--- Quote from: Eithanius on 27 Apr 2009, 04:07:25 am ---Were there any performance enhancements with this mod compared to stock Cuda client other than to consolidate the WUs to both CPUs and GPUs...? It seems my GPU didn't get any performance boost at all.
Not that I'm underestimating Raistmer's work, I do appreciate it.
Anyhow, my daily quota has been reduced from 1200 to 500... Any remedy to this..?
--- End quote ---
Yes, current CUDA MB build has advantages in speed versus stock one. Not so big as our other opt apps have though.
Dayly quota can be reduced due pack of VLARs rejected by CUDA MB for example. IF this is a reason you could try to use rebranding script to move VLAR tasks from GPU to CPU queue.
k6xt:
--- Quote from: Richard Haselgrove on 27 Apr 2009, 12:23:50 pm ---
--- Quote from: k6xt on 27 Apr 2009, 11:56:44 am ---I'm successfully processing only AP and CUDA WU on a pair of Q6600 with CUDA cards using the optimized apps. Over the past few weeks my RAC has doubled to about 16,000 mostly due to these two PCs. All seems to be working fine except:
The stderr files shows CUDA's are taking on the order 100 seconds. But the calendar time seems to be a few hours. The CUDA cards run about as hot as usually stated on this site. I'm hoping an expert can spot why the CUDA's seem to take up excessive calendar time. Appreciate any help!
Here is the stderr for a typical work unit:
http://setiathome.berkeley.edu/result.php?resultid=1208947940
and for the same WU this is the BOINC Message showing almost 2 hours to complete:
2009-04-27 05:00:25 SETI@home Starting task 01fe09aa.6562.18886.6.8.69_0 using setiathome_enhanced version 608
2009-04-27 06:56:12 SETI@home Computation for task 01fe09aa.6562.18886.6.8.69_0 finished
At these times the PC was idle and, in any case, I permit CUDA computation while the PC is in use.
--- End quote ---
The line in stderr says "CPU time 139.5781"
That's how long the Q6600 spent on them. Your 8500 GT (unfortunately, one of the slowest CUDA cards available) may indeed have taken almost two hours to complete the task. But during that time, all but a minute fraction of the power of one core of the Q6600 was available for AP, browsing the web, or whatever. So the CUDA contribution, comparatively small as it is, is all a net gain to your productivity.
--- End quote ---
Re CUDA processing time on 8500: Understood. Its GFLOPS rating is <10, my 9600GTX+ is 100. Plus the 8500 is the very first - oldest - on the NVIDIA CUDA list. It was my impression that time is in seconds, so the card processed for 139 seconds over a 2 hour period. Correct?
Thanks - Art
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version