Advanced search

Message boards : Graphics cards (GPUs) : WU not starting

Author Message
Jamie Kotlewski
Send message
Joined: 5 Aug 09
Posts: 2
Credit: 55,969
RAC: 0
Level

Scientific publications
wat
Message 14977 - Posted: 4 Feb 2010 | 1:22:31 UTC

Hey everyone,

Just been trying to run the new Beta application to test out the speed increase. However it doesn't start. I usually run Collatz on my computer. (MacbookPro running Win7) Boinc 6.10.29

The Collatz runs fine but when there aren't any Collatz WU's and only GPUGRID is available it doesn't start crunching, only saying "Ready to Start". I tried resetting the project, closing and opening Boinc etc. Any help would be appreciated as I would like to contribute.

Thanks in advance.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 14980 - Posted: 4 Feb 2010 | 8:37:07 UTC - in response to Message 14977.

I had the same problem on 2 machines. First try rebooting. If that doesn't work, reset the project. If that doesn't work, detach and reattach.

Jamie Kotlewski
Send message
Joined: 5 Aug 09
Posts: 2
Credit: 55,969
RAC: 0
Level

Scientific publications
wat
Message 14998 - Posted: 4 Feb 2010 | 23:00:54 UTC - in response to Message 14980.

Just figured out that I need a video card with 384 megs of memory, That error didn't pop up until recently. Shoot.

Thanks though!

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15001 - Posted: 5 Feb 2010 | 4:24:57 UTC - in response to Message 14998.

I also have a card that now won't start GPUGRID WUs. It's been running this project for months and all of sudden will DL WUs but won't start them. Other GPU projects run fine. It has 384MB memory, all of my other cards have 512MB or more. Has the memory requirement been raised in the last week or so?

Toni
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 9 Dec 08
Posts: 1006
Credit: 5,068,599
RAC: 0
Level
Ser
Scientific publications
watwatwatwat
Message 15012 - Posted: 5 Feb 2010 | 11:02:20 UTC - in response to Message 15001.
Last modified: 5 Feb 2010 | 11:05:43 UTC

Does this happen with beta wus only? Beyond: which host please?

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15025 - Posted: 5 Feb 2010 | 16:20:24 UTC - in response to Message 15012.

Does this happen with beta wus only? Beyond: which host please?

Good question and one I've been trying to remember. I believe it started when v6.05 was released. It would DL WUs but wouldn't start them. I rebooted, aborted the WUs, reset the project, reinstalled BOINC, detached and re-attached. Finally it started running the WUs again. It ran a few more WUs then DLed a couple v6.08 and refused to start them. At that point I rebooted, aborted the WUs, reset the project, reinstalled BOINC, detached and re-attached twice. Nothing, so I swapped in a GT 240 that was in a different box. After swapping in the GT 240 the WUs immediately started to run and have been running ever since. This is the host:

http://www.gpugrid.net/show_host_detail.php?hostid=55407

The card was a 9600GSO/384MB that's been running great for a long time. The one WU error on 1/27 was due to the fact that I forgot to pause the client while playing a video, my fault.

Win7-64, NVidia v195.62, BOINC v6.29 (tried other versions too, even v6.32). My other 2 9600GSO cards did not have the problem, but have => 512MB of memory. If I may hazard a guess: is the client set to need 384MB memory? If so it needs to be set to something lower, even 383MB might do the trick. Strange thing is that it did run a v6.06 and 3 v6.08 WUs before it quit starting them altogether.

Toni
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 9 Dec 08
Posts: 1006
Credit: 5,068,599
RAC: 0
Level
Ser
Scientific publications
watwatwatwat
Message 15032 - Posted: 5 Feb 2010 | 21:54:59 UTC - in response to Message 15025.
Last modified: 6 Feb 2010 | 1:05:30 UTC

Thanks for the detailed report. It could be that at some point a server update raised the CUDA memory requirement by a tiny bit (the official scheduler has undergone many changes lately).

However, right now there are WUs being crunched by 9600 GTO's reporting 402653184 bytes (=384MB), e.g. 1814132. Nevertheless, video memory may be eroding the amount a bit, or the client may be misreporting.

More 384MB crunchers having/not having the same problem?

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15034 - Posted: 6 Feb 2010 | 4:12:37 UTC - in response to Message 15032.

Thanks for the detailed report. It could be that at some point a server update raised the CUDA memory requirement by a tiny bit (the official scheduler has undergone many changes lately).

However, right now there are WUs being crunched by 9600 GTO's reporting 402653184 bytes (=384MB), e.g. 1814132. Nevertheless, video memory may be eroding the amount a bit, or the client may be misreporting.

More 384MB crunchers having/not having the same problem?

The machine you linked hasn't run any beta WUs:

http://www.gpugrid.net/results.php?hostid=62235

I don't think the old WUs are a problem. How much memory does the beta app require? Can't it be set to ask for a little less than 384MB? They had the same problem for a day or two at Collatz a while back. The client was set to ask for more memory than it actually needed. A quick update of the code and even 256MB cards were good as the Collatz app only really used less than 200.

Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 1576
Credit: 5,605,461,851
RAC: 8,706,681
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15037 - Posted: 6 Feb 2010 | 12:11:09 UTC

My understanding is that the default BOINC server code - supplied as an example for projects to customise only - suggests:

Minimum GPU RAM 256 MB for plan_class cuda
Minimum GPU RAM 384 MB for plan_class cuda23

(perhaps minus a couple of meg in both cases to allow for rounding errors in client reporting). If your GPU has less than the minimum specified in the plan_class, the scheduler shouldn't send you any work in the first place.

In addition, there's new code in BOINC clients v6.10.25 and later, which checks the amount of available GPU RAM, and won't start the task if that's below the minimum specified by the project.

This bites if you have an operating system with fancy video effects, like Windows Vista/7, OS X, or even some of the fancier Linux desktops. [Some people have said that merely setting high-resolution display modes can cause problems, but I don't think that's confirmed, or even likely].

If those conditions describe your setup, try turning off the OS graphics effects (and any other program that might be using GPU RAM), and try again.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15038 - Posted: 6 Feb 2010 | 13:52:05 UTC - in response to Message 15037.

My understanding is that the default BOINC server code - supplied as an example for projects to customise only - suggests:

Minimum GPU RAM 256 MB for plan_class cuda
Minimum GPU RAM 384 MB for plan_class cuda23

(perhaps minus a couple of meg in both cases to allow for rounding errors in client reporting). If your GPU has less than the minimum specified in the plan_class, the scheduler shouldn't send you any work in the first place.

In addition, there's new code in BOINC clients v6.10.25 and later, which checks the amount of available GPU RAM, and won't start the task if that's below the minimum specified by the project.

This bites if you have an operating system with fancy video effects, like Windows Vista/7, OS X, or even some of the fancier Linux desktops. [Some people have said that merely setting high-resolution display modes can cause problems, but I don't think that's confirmed, or even likely].

If those conditions describe your setup, try turning off the OS graphics effects (and any other program that might be using GPU RAM), and try again.

Thanks Richard. System is Win7 but Aero is turned off and resolution is set to 768x1024. No screensaver, plain as can be. No other programs running, just a dedicated cruncher box. Does GPUGRID need to specify 384MB ram or is it that they're just using the default? As I mentioned above the same problem happened at Collatz and they just changed the code to specify less ram since their client didn't use anywhere near that much anyway.

Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 1576
Credit: 5,605,461,851
RAC: 8,706,681
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15039 - Posted: 6 Feb 2010 | 14:11:58 UTC - in response to Message 15038.

Thanks Richard. System is Win7 but Aero is turned off and resolution is set to 768x1024. No screensaver, plain as can be. No other programs running, just a dedicated cruncher box. Does GPUGRID need to specify 384MB ram or is it that they're just using the default? As I mentioned above the same problem happened at Collatz and they just changed the code to specify less ram since their client didn't use anywhere near that much anyway.

GPUGrid seem to have set

<gpu_ram>249561088.000000</gpu_ram>

for the new Beta app (<plan_class> cuda), but not set any limit for the established acemd v6.71 (cuda or cuda23)

I see your host 55407 has run two betas overnight: maybe they set that 238MB limit after your post.

Toni
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 9 Dec 08
Posts: 1006
Credit: 5,068,599
RAC: 0
Level
Ser
Scientific publications
watwatwatwat
Message 15041 - Posted: 6 Feb 2010 | 14:44:25 UTC - in response to Message 15039.

We didn't explicitly change the setting, but I'm glad it started working.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15045 - Posted: 7 Feb 2010 | 0:03:02 UTC - in response to Message 15039.

<gpu_ram>249561088.000000</gpu_ram>

for the new Beta app (<plan_class> cuda), but not set any limit for the established acemd v6.71 (cuda or cuda23)

I see your host 55407 has run two betas overnight: maybe they set that 238MB limit after your post.

Not fixed. The only reason the host ran those WUs is because I swapped in a GT 240 from a different machine.
Looks like your info above may be a clue to the solution though.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 15046 - Posted: 7 Feb 2010 | 0:04:06 UTC - in response to Message 15041.
Last modified: 7 Feb 2010 | 0:04:40 UTC

We didn't explicitly change the setting, but I'm glad it started working.

It didn't. See the post just above.

Profile mike047
Send message
Joined: 21 Dec 08
Posts: 47
Credit: 7,330,049
RAC: 0
Level
Ser
Scientific publications
watwatwatwatwatwatwat
Message 15234 - Posted: 15 Feb 2010 | 13:17:58 UTC
Last modified: 15 Feb 2010 | 13:20:22 UTC

I am having a similar "no start" issue. GTX260 with 196.34 drivers and 6.10.18 boinc client.

Any help??

Edit; the issue was with the boinc preferences....I am good to go.
____________
mike

Post to thread

Message boards : Graphics cards (GPUs) : WU not starting

//