nvidia GPU - 8.22 (opencl_nvidia_SoG) issues

Questions and Answers : GPU applications : nvidia GPU - 8.22 (opencl_nvidia_SoG) issues
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Chris Feyen

Send message
Joined: 1 Jan 01
Posts: 20
Credit: 4,652,532
RAC: 15,048
United States
Message 1850048 - Posted: 19 Feb 2017, 23:16:30 UTC

I have a few Ubuntu computers with an Nvidia video card in them. BOINC gets about 30 seconds into processing a task, then moves on to a different task for the video card. The original task goes to postponed status, then waiting. If a computer only has 3 tasks in the queue for the video card and all 3 tasks are application 8.22 (opencl_nvidia_SoG), it will bounce between all 3 tasks.

Is it possible to block 8.22 (opencl_nvidia_SoG) tasks from being downloaded? I'm not sure of what other options are available.

Thanks,
Chris
~Chris

ID: 1850048 · Report as offensive
Profile Tom Miller
Volunteer tester
Avatar

Send message
Joined: 28 Nov 02
Posts: 766
Credit: 16,683,165
RAC: 45,701
United States
Message 1856394 - Posted: 18 Mar 2017, 18:44:16 UTC - in response to Message 1850048.  

I have a few Ubuntu computers with an Nvidia video card in them. BOINC gets about 30 seconds into processing a task, then moves on to a different task for the video card. The original task goes to postponed status, then waiting. If a computer only has 3 tasks in the queue for the video card and all 3 tasks are application 8.22 (opencl_nvidia_SoG), it will bounce between all 3 tasks.

Is it possible to block 8.22 (opencl_nvidia_SoG) tasks from being downloaded?


I know that Bonic will switch between near alike tasks (in the names) from different projects (like Seti and Seti beta). Is that what is happening for you?

I don't know if you can "block" (or simply not choose them) those tasks in "stock" seti but I believe you can in the Lunatics (http://lunatics.kwsn.info/index.php) version.

If you want to shutdown all gpu tasks you can simply define the Ubuntu computers to be in a different group from you main group (home vs. work vs. school) and tell that group "no gpu" tasks.

HTH,
Tom
"You are entitled to your own opinion but not to your own facts." Senator and Professor Patrick Moynihan
---
https://GalensonConsulting.WordPress.com
ID: 1856394 · Report as offensive
Kissagogo27Project Donor

Send message
Joined: 6 Nov 99
Posts: 52
Credit: 3,601,669
RAC: 4,865
France
Message 1856538 - Posted: 19 Mar 2017, 10:32:09 UTC

Hi, you could have some answers by looking at the Stderr.txt file in the Slots directories ( sub folders 0 1 n ... depend of how many tasks you're running at the same time ) in the Boinc Data folder to know what is the error make them postponned ...

do you reserve one core by gpu task because of Sog tasks needs more CPU activity ?
ID: 1856538 · Report as offensive
Profile Tom Miller
Volunteer tester
Avatar

Send message
Joined: 28 Nov 02
Posts: 766
Credit: 16,683,165
RAC: 45,701
United States
Message 1870226 - Posted: 31 May 2017, 0:38:57 UTC

I recently read that the symptoms you are experiencing can be caused by having too little ram on the video card compared to what you are trying to run.

What is not clear is does that mean the -sbs # is too large (in the MB*SOG.txt).

Or something else. In any case, the -sbs needs to be less the full ram of your gpu. I have seen numbers as large as half the hardware listing of ram in the gpu. It could be as little as -sbs 192 or less.

HTH,

Tom
"You are entitled to your own opinion but not to your own facts." Senator and Professor Patrick Moynihan
---
https://GalensonConsulting.WordPress.com
ID: 1870226 · Report as offensive
Profile Tom Miller
Volunteer tester
Avatar

Send message
Joined: 28 Nov 02
Posts: 766
Credit: 16,683,165
RAC: 45,701
United States
Message 1870248 - Posted: 31 May 2017, 2:20:46 UTC

I see a couple of Linux boxes with GT 710's on them. The gpu's have 192 Cuda cores as well as a base clock of 954 Mhz. So they may be taking (at best) 1.5 hours or more per gpu-processed task.

I believe I ran that video card under Windows 7. And it worked without any back talk. So, it sounds like it is a Linux/Seti driver or setup issue.

There is a thread about setting up a Linux box from scratch to run the super-fast cuda80 beta app. I think it is in the "Number Crunching" message area. Maybe that will help.

Tom
"You are entitled to your own opinion but not to your own facts." Senator and Professor Patrick Moynihan
---
https://GalensonConsulting.WordPress.com
ID: 1870248 · Report as offensive
Profile Mario Lampe

Send message
Joined: 9 Apr 17
Posts: 3
Credit: 143,078
RAC: 48
Germany
Message 1870855 - Posted: 3 Jun 2017, 7:59:23 UTC

Hi,

is there any solution or workaround for this problem ? I have the same issue with Debian Linux 8.8, BOINC 7.6.33 and nVidia GT 710, all other WUs working fine (cudaXX, sah) - only SoG-WUs stopping the calculation with the following error in stderr.txt:

 ERROR: OpenCL kernel/call 'clGetEventProfilingInfo' call failed (-7) in file ../../src/GPU_lock.cpp near line 550


Thanks for any help.

Mario
ID: 1870855 · Report as offensive
Profile Ageless
Avatar

Send message
Joined: 9 Jun 99
Posts: 14183
Credit: 3,471,282
RAC: 1,498
Netherlands
Message 1870895 - Posted: 3 Jun 2017, 14:27:20 UTC - in response to Message 1870855.  

You can read through https://setiathome.berkeley.edu/forum_thread.php?id=80847 to see what the options are (delete the corresponding *.bin and *.wisdom files in the Seti projects directory, then reboot), or repost this in Number Crunching, where the developer hangs out.
Jord

Ancient Astronaut Theorists suggest that in many ways, you can be considered an alien conspiracy!
ID: 1870895 · Report as offensive

Questions and Answers : GPU applications : nvidia GPU - 8.22 (opencl_nvidia_SoG) issues


 
©2017 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.