Strange Issue

Wicked_Pixie said:

Thanks for the reply MD!
Well, I thought it was prioritisng due dates too. But, when I looked at the dates, it was totally random.
According to these threads here and here, It seems to be a BOINC issue - EDF specifically ( I don't know what that means).


I'm now running 6.10.4. Same issue as the last version, but, at least, I hardly get the display driver quitting on me.

However, I found out accidentally that if I disable SLI in NV control panel. It completes the WU very quickly (under 4Min.) Boinc only sees 1 GPU core though. Is it possible that 2 both GPU cores were running the WU at the same time? I dunno.
But if I can confirm later that the results are good, I may just disable SLI entirely.
Anyone with SLI cards care to try this for testing?

Click to expand...

Codeman05 said:

yes, you will need the dummy plug on the second GPU in order for windows to allow you to enable the ghost display so BOINC will see it.

Though one "trick" i noticed was IF you enable PhysX (no sli), then you can enable the second monitor with no dummy plug...not sure why that works, but I've been unable to do that under Vista and W7

Click to expand...

Wicked_Pixie said:

On one of my rigs, CUDA WU would start, pause 'waiting to run', and move on to a new WU.
I have now about a dozen WU's 'waiting to run' at various stages of work completed. When it resumes on WUs that are 99% complete, I get an error - display driver stopped working, and then picks up a new WU.

Not all WUs that resume from '99% done', end up as 'computational error', but they are increasing in frequency.

Never did that before until last night...

Using 190.62 drivers w/ CUDA 2.3 drop-in files and optimised unified installer.
GPU is GTX295, if that makes any difference.

Anyone had this problem?
Should I re-install driver or go back to 190.38??
Anyone using BOINC 6.10.4? Would that be better?

Click to expand...

Duner said:

This is a very common issue actually. There was a thread on it at the SETI forums, but I can't find it. I did find the solution threat though.

FIX

The problem is SETI doesn't account for the speed boost of the optimized apps and therefore the predicted end times are not valid, so they jump around all the time. By manually adding the flop count for the CPU and GPU, you somewhat mitigate the problem. I've applied the flops correction myself and haven't had units paused in the middle for a long time now.

The only part I wound change is:

8. For each of the apps multiply the p_fpops value by the factor below and put this into the appropiate flops entry in the app_info given below. For multibeam 608 you need the estimated Gflops. The app_info given below has the values for a GTS250.
Application Calculate
Astropulse 503 = p_fpops x 2.6
Astropulse 505 = P_fpops x 2.6
Multibeam 603 = p_fpops x 1.75
Multibeam 608 = Est.Gflops x 0.2

If you are using the latest unified installer, change the Est. Gflops to x 0.3 and if you are using the 2.3 dll drop in files, then use a x 0.4 or even 0.5 to account for the improved performance.

Click to expand...

Thanks for the tip and link, duner !!!

So, I looked at my app_info and this is what it looks like.

- <app_version>
  <app_name>astropulse_v505</app_name> 
  <version_num>505</version_num> 
- <file_ref>
  <file_name>ap_5.05r168_SSE3.exe</file_name> 
  <main_program /> 
  </file_ref>
  </app_version>
- <app>
  <name>setiathome_enhanced</name> 
  </app>
- <file_info>
  <name>MB_6.08_CUDA_V12_VLARKill_FPLim2048.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> 
  <plan_class>cuda</plan_class> 
  <avg_ncpus>0.040000</avg_ncpus> 
  <max_ncpus>0.040000</max_ncpus> 
- <coproc>
  <type>CUDA</type> 
  <count>1</count> 
  </coproc>
- <file_ref>
  <file_name>MB_6.08_CUDA_V12_VLARKill_FPLim2048.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_info>

Now, I searched for the client_state.xml file (to search for P_fpops figures) and cannot find it anywhere.
I looked at the Boinc files in Program files and the AppData files.

For 3593 floating point MIPS (Whetstone) per CPU, <flops>3593e+6</flops> is the simplest. 3.593e+9 or 3593000000 are other forms you could use.
Joe

Click to expand...

Duner, if it is not too much trouble, would it be alright for you to post or PM me your app_info? I just want to compare excatly where I insert them. I am afraid of borking that file. I'm in no hurry though, since I am finishing up all my queued WUs before editing the app_info file.

Thanks again for your help.

QuietIce said:

If I remember right (and I don't run SLI so this is something dredged from memory) you should set your cards to run independently, not in SLI. Can't remember how to get BOINC to see the other card(s) though. I'll try to find the thread where that was discussed - it's been awhile.


Did you reboot after you disabled SLI ...?


Edit:
Success! Know anything about "dummy plugs"?
http://www.ocforums.com/showpost.php?p=6092361&postcount=6
http://www.ocforums.com/showthread.php?t=605690

Click to expand...

Wicked_Pixie said:

Thanks for the tip and link, duner !!!

So, I looked at my app_info and this is what it looks like.

Code:

- <app_version>
  <app_name>astropulse_v505</app_name> 
  <version_num>505</version_num> 
- <file_ref>
  <file_name>ap_5.05r168_SSE3.exe</file_name> 
  <main_program /> 
  </file_ref>
  </app_version>
- <app>
  <name>setiathome_enhanced</name> 
  </app>
- <file_info>
  <name>MB_6.08_CUDA_V12_VLARKill_FPLim2048.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> 
  <plan_class>cuda</plan_class> 
  <avg_ncpus>0.040000</avg_ncpus> 
  <max_ncpus>0.040000</max_ncpus> 
[COLOR="Yellow"] <flops>xxxxxxxxxxx.x</flops>[/COLOR]
- <coproc>
  <type>CUDA</type> 
  <count>1</count> 
  </coproc>
- <file_ref>
  <file_name>MB_6.08_CUDA_V12_VLARKill_FPLim2048.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_info>

Now, I searched for the client_state.xml file (to search for P_fpops figures) and cannot find it anywhere.
I looked at the Boinc files in Program files and the AppData files.

The p_fpops for the GTX295 should be easy to figure out . But since math is not my strong suit, could someone see if this is correct...
106GFLOPS = 106000000000 flops x 0.5 = 53000000000 flops

So, I need to figure out what my CPU flops are.
According to this post, you can estimate it by using MIPS and add six zeroes behind it.


Based on that example, is that computation correct?

Duner, if it is not too much trouble, would it be alright for you to post or PM me your app_info? I just want to compare excatly where I insert them. I am afraid of borking that file. I'm in no hurry though, since I am finishing up all my queued WUs before editing the app_info file.

Thanks again for your help.

Click to expand...

SETI@Home Forums said:

4. Browse the BOINC log file to get the estimated speed of your GPU (or before you shut BOINC down, click on the messages tab). This is usually given at the top and is in Gflops. Some estimates from my testing are:
a) 9800GT = 60Gflops
b) GTS250 = 84Gflops
c) GTX260 (216 sp) = 96Gflops

Click to expand...

LandShark said:

while we at it, does boinc/seti benefit more from GPU core/memory speed or (like folding) from shader speed??

Click to expand...