Four work units cycling every 12 seconds

Questions and Answers : GPU applications : Four work units cycling every 12 seconds
Message board moderation

To post messages, you must log in.

AuthorMessage
George7

Send message
Joined: 29 Mar 04
Posts: 6
Credit: 3,052,728
RAC: 1
United Kingdom
Message 2014151 - Posted: 4 Oct 2019, 14:55:36 UTC

I had this trouble once before a few weeks ago and aborted the work units but it seems to have come back with a vengeance. I have a stock of WUs waiting to go through using the NVIDIA. One WU starts, processes for about 12 seconds then postpones. A second one starts, same thing. Then a third and fourth exactly the same. Then it cycles back to the first WU, restarts it, runs 12 seconds, postpones, on to the second WU again, runs 12 seconds, postpones, etc, etc, in an unending loop. Each time the clock is restarted so no progress is being made. I've tried rebooting to no effect. Also noted it's blanking the screen momentarily when running in the background. Any help truly appreciated.
ID: 2014151 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2014152 - Posted: 4 Oct 2019, 15:00:05 UTC - in response to Message 2014151.  

You have a missing OpenCL driver that the task needs to run. Check your Event Log.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2014152 · Report as offensive
George7

Send message
Joined: 29 Mar 04
Posts: 6
Credit: 3,052,728
RAC: 1
United Kingdom
Message 2014153 - Posted: 4 Oct 2019, 15:07:50 UTC - in response to Message 2014152.  

Thank you for you prompt reply. My knowledge of these systems is rather limited. I've had a look at the Event Log but it doesn't mention the OpenCL driver. Can it be rectified?
ID: 2014153 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2014157 - Posted: 4 Oct 2019, 15:25:56 UTC - in response to Message 2014153.  
Last modified: 4 Oct 2019, 15:28:01 UTC

Don't use Microsoft graphics drivers or let Micrsoft install or update the graphics drivers. Go to the Nvidia download site and download and install the 431 series drivers. Those are the last that are working correctly for BOINC. Later drivers have issues.
Once you install the new drivers and reboot the computer you should see the card being detected at startup in the Event Log. You should see a statement of the CUDA component being detected followed by a statement that the OpenCL component is detected. Seti needs a properly installed OpenCL component to run the gpu tasks.

You might want to read this thread which is specifically about the issue.
https://setiathome.berkeley.edu/forum_thread.php?id=84694
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2014157 · Report as offensive
George7

Send message
Joined: 29 Mar 04
Posts: 6
Credit: 3,052,728
RAC: 1
United Kingdom
Message 2014158 - Posted: 4 Oct 2019, 15:34:28 UTC - in response to Message 2014157.  

Thanks very much for this Keith. I'll give it a go and update you on the results in due course.
ID: 2014158 · Report as offensive
George7

Send message
Joined: 29 Mar 04
Posts: 6
Credit: 3,052,728
RAC: 1
United Kingdom
Message 2014183 - Posted: 4 Oct 2019, 21:42:40 UTC - in response to Message 2014158.  
Last modified: 4 Oct 2019, 22:00:11 UTC

I went to the Nvidia download site but was unable to find the 431 series drivers, only the latest version 436.48 which I downloaded and installed. Below is a copy of the log entry after the download.

04/10/2019 22:03:21 | | CUDA: NVIDIA GPU 0: GeForce GTX 980 (driver version 436.48, CUDA version 10.1, compute capability 5.2, 4096MB, 3378MB available, 5497 GFLOPS peak)
04/10/2019 22:03:21 | | OpenCL: NVIDIA GPU 0: GeForce GTX 980 (driver version 436.48, device version OpenCL 1.2 CUDA, 4096MB, 3378MB available, 5497 GFLOPS peak)

However it has made no difference unfortunately. Having said this, I have noted that it will process satisfactorily the WUs tagged with "vlar_1" at the end of the description line. These WUs have an estimated runtime of 10:03. The ones without this tag have an estimated runtime of 3:52 and it is these (unfortunately the majority) which are giving the postpone-reset problem.

I will re-visit the Nvidia site to try to find the 431 series and let you know how I get on.

EDIT UPDATE - I have just hit the project reset which has effectively cleared the stock of mainly non-vlar WUs. I have now downloaded a day's worth of new WUs which, apart from one or two, are all tagged with "vlar_1" or "vlar_0". The system seems quite happy now to munch its way through these. Hopefully the SETI download system recognises my computer's dietary restrictions.

However, I will continue to pursue the 431 series drivers as per your advice. Many thanks again, George7.
ID: 2014183 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2014187 - Posted: 4 Oct 2019, 22:00:00 UTC - in response to Message 2014183.  

You have to go to the Beta and Legacy drivers link.
https://www.nvidia.com/download/driverResults.aspx/148865/en-us
will get you to the Windows 10 64 bit 431.60 drivers.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2014187 · Report as offensive
George7

Send message
Joined: 29 Mar 04
Posts: 6
Credit: 3,052,728
RAC: 1
United Kingdom
Message 2014189 - Posted: 4 Oct 2019, 22:01:11 UTC - in response to Message 2014187.  
Last modified: 4 Oct 2019, 22:01:46 UTC

Thanks Keith, will do. (I have edited my post above).
ID: 2014189 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2014190 - Posted: 4 Oct 2019, 22:02:14 UTC - in response to Message 2014183.  

I went to the Nvidia download site but was unable to find the 431 series drivers, only the latest version 436.48 which I downloaded and installed. Below is a copy of the log entry after the download.

04/10/2019 22:03:21 | | CUDA: NVIDIA GPU 0: GeForce GTX 980 (driver version 436.48, CUDA version 10.1, compute capability 5.2, 4096MB, 3378MB available, 5497 GFLOPS peak)
04/10/2019 22:03:21 | | OpenCL: NVIDIA GPU 0: GeForce GTX 980 (driver version 436.48, device version OpenCL 1.2 CUDA, 4096MB, 3378MB available, 5497 GFLOPS peak)

However it has made no difference unfortunately. Having said this, I have noted that it will process satisfactorily the WUs tagged with "vlar_1" at the end of the description line. These WUs have an estimated runtime of 10:03. The ones without this tag have an estimated runtime of 3:52 and it is these (unfortunately the majority) which are giving the postpone-reset problem.

I will re-visit the Nvidia site to try to find the 431 series and let you know how I get on.

EDIT UPDATE - I have just hit the project reset which has effectively cleared the stock of mainly non-vlar WUs. I have now downloaded a day's worth of new WUs which, apart from one or two, are all tagged with "vlar_1" or "vlar_0". The system seems quite happy now to munch its way through these. Hopefully the SETI download system recognises my computer's dietary restrictions.

However, I will continue to pursue the 431 series drivers as per your advice. Many thanks again, George7.

Yes, the BLC VLAR's will process correctly. The Arecibo ones are hit and miss, mostly miss because of their angle ranges. The 431 drivers process all the tasks correctly. The problem isn't just with Seti either. Other projects are having similar problems with certain of their work units too.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2014190 · Report as offensive
George7

Send message
Joined: 29 Mar 04
Posts: 6
Credit: 3,052,728
RAC: 1
United Kingdom
Message 2014192 - Posted: 4 Oct 2019, 22:14:53 UTC - in response to Message 2014190.  

04/10/2019 23:11:23 | | CUDA: NVIDIA GPU 0: GeForce GTX 980 (driver version 431.60, CUDA version 10.1, compute capability 5.2, 4096MB, 3378MB available, 5497 GFLOPS peak)
04/10/2019 23:11:23 | | OpenCL: NVIDIA GPU 0: GeForce GTX 980 (driver version 431.60, device version OpenCL 1.2 CUDA, 4096MB, 3378MB available, 5497 GFLOPS peak)

There we go! Thanks very much for that Keith, and the extra info. As I said, I'll keep this thread on a shortcut for instant access if I need it again. George7.
ID: 2014192 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 2014193 - Posted: 4 Oct 2019, 22:27:49 UTC - in response to Message 2014192.  

04/10/2019 23:11:23 | | CUDA: NVIDIA GPU 0: GeForce GTX 980 (driver version 431.60, CUDA version 10.1, compute capability 5.2, 4096MB, 3378MB available, 5497 GFLOPS peak)
04/10/2019 23:11:23 | | OpenCL: NVIDIA GPU 0: GeForce GTX 980 (driver version 431.60, device version OpenCL 1.2 CUDA, 4096MB, 3378MB available, 5497 GFLOPS peak)

There we go! Thanks very much for that Keith, and the extra info. As I said, I'll keep this thread on a shortcut for instant access if I need it again. George7.

Great news. You shouldn't have any issues with tasks stalling out now. Those drivers are up to date with all the latest security mitigations in place. Generally, we don't need to update graphics drivers when running compute. The drivers from 2 years ago work fine. The only reason to ever update graphics drivers is when a new family of cards are introduced which need the latest drivers to recognize them. The other reason is if you are a gamer and have major issues with your current drivers on certain games and that bug you so much you need to update. But that comes with the possibility of breaking compute as you found out.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 2014193 · Report as offensive

Questions and Answers : GPU applications : Four work units cycling every 12 seconds


 
©2024 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.