Seti@Home optimized science apps and information

Optimized Seti@Home apps => Windows => GPU crunching => Topic started by: Skywalker66_Bln on 07 Apr 2010, 11:55:01 am

Title: Problem long crunch time
Post by: Skywalker66_Bln on 07 Apr 2010, 11:55:01 am

i have a problem on one of my computers,
GPU crunching time is normal 0:20-1:30h, in the last days i have a few WUs that run 3:30. that is not the problem. Because all wu-times of uncrunched were crash to high times. 

WU http://setiathome.berkeley.edu/workunit.php?wuid=594291916 as example
WU true angle range is :  0.294035
it is no Vlar
cpu usage by crunching is by low 0,0015-0,0081%

boinc 6.10.43 / Cuda 2.3
GPU are calculateted with 131gflops peak /5,6 x0,5 = 11,696 gflops ( on my other Computers run this calculation perfectly)
Cpu client_state-xml for <p_flops> x1,75

  <app>
  <name>setiathome_enhanced</name>
  </app>
   
   <file_info>
   <name>MB_6.08_CUDA_V12_VLARKill_FPLim2048_test.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>608</version_num>
   <platform>windows_intelx86</platform>
   <plan_class>cuda</plan_class>
   <avg_ncpus>0.040000</avg_ncpus>
   <max_ncpus>0.040000</max_ncpus>
   <flops>11696428571,428600</flops>
   <coproc>
     <type>CUDA</type>
     <count>1</count>
   </coproc>
   <file_ref>
   <file_name>MB_6.08_CUDA_V12_VLARKill_FPLim2048_test.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>
  </app_version>
   
  <app_version>
   <app_name>setiathome_enhanced</app_name>
   <version_num>608</version_num>
   <platform>windows_intelx86</platform>
   <plan_class>cuda23</plan_class>
   <avg_ncpus>0.040000</avg_ncpus>
   <max_ncpus>0.040000</max_ncpus>
   <flops>11696428571,428600</flops>
   <coproc>
     <type>CUDA</type>
     <count>1</count>
   </coproc>
   <file_ref>
   <file_name>MB_6.08_CUDA_V12_VLARKill_FPLim2048_test.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>
  </app_version>
   
  <app_version>
   <app_name>setiathome_enhanced</app_name>
   <version_num>609</version_num>
   <platform>windows_intelx86</platform>
   <plan_class>cuda</plan_class>
   <avg_ncpus>0.040000</avg_ncpus>
   <max_ncpus>0.040000</max_ncpus>
   <flops>11696428571,428600</flops>
   <coproc>
     <type>CUDA</type>
     <count>1</count>
   </coproc>
   <file_ref>
   <file_name>MB_6.08_CUDA_V12_VLARKill_FPLim2048_test.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>
  </app_version>   

  <app_version>
   <app_name>setiathome_enhanced</app_name>
   <version_num>609</version_num>
   <platform>windows_intelx86</platform>
   <plan_class>cuda23</plan_class>
   <avg_ncpus>0.040000</avg_ncpus>
   <max_ncpus>0.040000</max_ncpus>
   <flops>11696428571,428600</flops>
   <coproc>
     <type>CUDA</type>
     <count>1</count>
   </coproc>
   <file_ref>
   <file_name>MB_6.08_CUDA_V12_VLARKill_FPLim2048_test.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>
  </app_version>
 

Have anyone a good idea ? or a better calculation ?

Thanks for help

Title: Re: Problem long crunch time
Post by: Skywalker66_Bln on 07 Apr 2010, 05:37:32 pm

i am back to the basics

cpu <p_flops> x0
gpu Gflops peak /5,6 (to east. gflops) then *0,2


post like http://setiathome.berkeley.edu/forum_thread.php?id=54801 make me crazy

it is a pity which has struck nobody here a little bit and give a comment in addition hand in has. Or only of all malicious pleasure?

I thought I am with the best ones of the best who also help each other...

Title: Re: Problem long crunch time
Post by: Claggy on 07 Apr 2010, 06:43:09 pm

Quote from: Skywalker66_Bln on 07 Apr 2010, 11:55:01 am

i have a problem on one of my computers,
GPU crunching time is normal 0:20-1:30h, in the last days i have a few WUs that run 3:30. that is not the problem. Because all wu-times of uncrunched were crash to high times. 

WU http://setiathome.berkeley.edu/workunit.php?wuid=594291916 as example
WU true angle range is :  0.294035
it is no Vlar
cpu usage by crunching is by low 0,0015-0,0081%

boinc 6.10.43 / Cuda 2.3
GPU are calculateted with 131gflops peak /5,6 x0,5 = 11,696 gflops ( on my other Computers run this calculation perfectly)
Cpu client_state-xml for <p_flops> x1,75

Snip

Have anyone a good idea ? or a better calculation ?

Thanks for help

I wouldn't worry about it, the Workunit AR is below a normal AR, but above the value for being a VLAR Wu, so wouldn't get aborted by the app,
as it happens, my 9800GTX+ has done some of a similiar AR, and took twice as long as a normal AR Wu:

http://setiathome.berkeley.edu/result.php?resultid=1554900415 (http://setiathome.berkeley.edu/result.php?resultid=1554900415)

http://setiathome.berkeley.edu/result.php?resultid=1554900210 (http://setiathome.berkeley.edu/result.php?resultid=1554900210)

http://setiathome.berkeley.edu/result.php?resultid=1554900203 (http://setiathome.berkeley.edu/result.php?resultid=1554900203)

http://setiathome.berkeley.edu/result.php?resultid=1554900180 (http://setiathome.berkeley.edu/result.php?resultid=1554900180)

Claggy

Title: Re: Problem long crunch time
Post by: Skywalker66_Bln on 08 Apr 2010, 06:27:12 am

I think that

gpu Gflops peak /5,6 (to east. gflops) then x0,5 (x0,09) is to high

i work on a simple optimisation way that can newbee´s or non englisch speaking people understand

at the moment, i am by a factor from x0,43 (x0,077). it shows as a pointlandet place. i am interested to show on my other Computers