Advanced search

Message boards : Graphics cards (GPUs) : GTX295 running two simultaneous WUs but only one WU is actually being computed

Author Message
frigens
Send message
Joined: 24 Feb 09
Posts: 14
Credit: 107,808,533
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwat
Message 8331 - Posted: 10 Apr 2009 | 1:48:42 UTC

I've been trying to get my GTX295 to work two tasks at the same time and have tried many config. I'm now running BOINC on WinXP 32-bit with driver version 182.50. SLI is off and PhysX is enabled.

BOINC can now run two CUDA WUs but it's really only one WU get processed by GTX295 and you can see that remaining time is decreasing. The other WU would remain at 0.0% like it's not running at all. Was there anything I missed?

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 8334 - Posted: 10 Apr 2009 | 5:37:11 UTC

I saw this problem with 6.6.20 and did not figure out what was happening ... so I went back to 6.5.0 ... we were trying to figure this out and I was hoping to get enough up so I could tackle looking at this again ... today was a bust so I will shoot for tomorrow ...

At first we thought it might be some bad tasks from the project. But, NOW, the possibilities of BOINC not handling multiple GPUs correctly may be easier to prove...

Try 6.5.0 to see if that gets the tasks going again ... or to the last version you had installed that worked well ... if it was higher than 6.5.0 then maybe we can point to the place where the bug got introduced.

My "best" suspicion is that BOINC is trying to run the tasks all one core. Can you check your temps before you make a change and see if one or more of your cores is not running?

Profile Dieter Matuschek
Avatar
Send message
Joined: 28 Dec 08
Posts: 58
Credit: 231,884,297
RAC: 0
Level
Leu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 8338 - Posted: 10 Apr 2009 | 10:32:01 UTC - in response to Message 8331.

I had this problem several times up to now - independent of the BOINC version.

Restarting BOINC was the solution.
Do you want to try it?
____________

frigens
Send message
Joined: 24 Feb 09
Posts: 14
Credit: 107,808,533
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwat
Message 8341 - Posted: 10 Apr 2009 | 13:24:04 UTC - in response to Message 8334.

I saw this problem with 6.6.20 and did not figure out what was happening ... so I went back to 6.5.0 ... we were trying to figure this out and I was hoping to get enough up so I could tackle looking at this again ... today was a bust so I will shoot for tomorrow ...

At first we thought it might be some bad tasks from the project. But, NOW, the possibilities of BOINC not handling multiple GPUs correctly may be easier to prove...

Try 6.5.0 to see if that gets the tasks going again ... or to the last version you had installed that worked well ... if it was higher than 6.5.0 then maybe we can point to the place where the bug got introduced.

My "best" suspicion is that BOINC is trying to run the tasks all one core. Can you check your temps before you make a change and see if one or more of your cores is not running?


I'm switching to version 6.4.7 and will see if it works. Couldn't find version 6.5.0 anywhere. If not work, maybe I should try version 6.6.21?

Profile asb2106
Send message
Joined: 24 Dec 08
Posts: 17
Credit: 10,751,419
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwat
Message 8347 - Posted: 10 Apr 2009 | 17:12:46 UTC - in response to Message 8341.
Last modified: 10 Apr 2009 | 17:14:05 UTC

6.5.0

I have been running 6.5.0 for atleast a couple months now, no problems, no reason to change...

to find all the versions go to http://boinc.berkeley.edu/dl

Snow Crash
Send message
Joined: 4 Apr 09
Posts: 450
Credit: 539,316,349
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 8349 - Posted: 10 Apr 2009 | 17:47:03 UTC

I am running BOINC 6.6.20 on Vista 64 and have two simultaneous WUs processing correctly. I only ever had problems when I first started a few days ago and did not have SLI off and PhysX on. Let me know of any other settings I can give you details on that may help you back to runing the max.

Steve

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 8353 - Posted: 10 Apr 2009 | 18:33:27 UTC

This may be a subtle and nasty bug. Some of the tasks on my system ran,or appeared to run correctly. Others took up to four times the normal time to run to completion. This may or may not be related to having up to 4 GPU processing elements in the system.

So, what *I* saw was a somewhat random walk of tasks with my normal 4 in work some of them would seem to run right, others incorrectly. Which is why I was looking for a GPU monitoring tool. I was wondering if BOINC is not using the right "pointer" to control the tasks in-flight.

As far as I can gather poor old Rom is using a dual core probably with a low end GPU to do development. What that means is that issues like the one I am describing on the mailing lists (and others) dealing with the CPU scheduler that start to show up on 4 core systems and is quite noticible on 8 CPU systems, well, he can't see them at all ...

frigens
Send message
Joined: 24 Feb 09
Posts: 14
Credit: 107,808,533
RAC: 0
Level
Cys
Scientific publications
watwatwatwatwatwatwatwat
Message 8357 - Posted: 10 Apr 2009 | 20:41:35 UTC

Thanks all for the comments, I'm giving BOINC v6.5.0 a try and I'll check out how it works out. Does NVIDIA driver has any roles in this problem? If so, which driver version is I should try to switch to? I'm using latest stable driver v182.50, is it a good version?

Profile Paul D. Buck
Send message
Joined: 9 Jun 08
Posts: 1050
Credit: 37,321,185
RAC: 0
Level
Val
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 8369 - Posted: 11 Apr 2009 | 4:17:22 UTC - in response to Message 8357.

Thanks all for the comments, I'm giving BOINC v6.5.0 a try and I'll check out how it works out. Does NVIDIA driver has any roles in this problem? If so, which driver version is I should try to switch to? I'm using latest stable driver v182.50, is it a good version?

On XP Pro 32-Bit that combination is what I have settled in on ... there is another earlier driver that is almost as good (180.22?, I can go look for it if you need it - though this one caused BSOD when I used the nvidia control panel).

Post to thread

Message boards : Graphics cards (GPUs) : GTX295 running two simultaneous WUs but only one WU is actually being computed

//