Advanced search

Message boards : Graphics cards (GPUs) : no new WU's, video RAM message

Author Message
cstaffa
Send message
Joined: 2 Aug 11
Posts: 5
Credit: 737,357
RAC: 0
Level
Gly
Scientific publications
watwatwatwat
Message 22848 - Posted: 27 Dec 2011 | 19:56:11 UTC

12/27/2011 10:13:12 AM GPUGRID Sending scheduler request: To fetch work.
12/27/2011 10:13:12 AM GPUGRID Requesting new tasks for GPU
12/27/2011 10:13:15 AM GPUGRID Scheduler request completed: got 0 new tasks
12/27/2011 10:13:15 AM GPUGRID Message from server: A minimum of 384 MB (preferably 384 MB) of video RAM is needed to process tasks using your computer's NVIDIA GPU
12/27/2011 10:13:15 AM GPUGRID Message from server: No tasks sent


I have not changed anything since my last WU two weeks ago. I do have two GPUs, only one of which is in use, and which does have 384MB of memory.

12/19/2011 1:38:23 PM NVIDIA GPU 0: GeForce 9600 GSO (driver version 26658, CUDA version 3020, compute capability 1.1, 384MB, 288 GFLOPS peak)
12/19/2011 1:38:23 PM NVIDIA GPU 1 (ignored by config): GeForce 9400 GT (driver version 26658, CUDA version 3020, compute capability 1.1, 1024MB, 43 GFLOPS peak)

____________

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 22850 - Posted: 27 Dec 2011 | 21:07:53 UTC - in response to Message 22848.

Try restarting your system.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

cstaffa
Send message
Joined: 2 Aug 11
Posts: 5
Credit: 737,357
RAC: 0
Level
Gly
Scientific publications
watwatwatwat
Message 22855 - Posted: 28 Dec 2011 | 19:02:37 UTC

I've restarted, and reset the project via BOINC Manager. I'm still getting the same message, and no new tasks.
____________

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 22857 - Posted: 28 Dec 2011 | 22:34:17 UTC - in response to Message 22855.
Last modified: 28 Dec 2011 | 23:51:28 UTC

I think this problem resulted from the recent server update.
The server software now appears to have a minimum requirement of 238MB for normal task types, and 384MB for long tasks.

Try selecting to only run normal length tasks (deselect long tasks).
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

cstaffa
Send message
Joined: 2 Aug 11
Posts: 5
Credit: 737,357
RAC: 0
Level
Gly
Scientific publications
watwatwatwat
Message 22858 - Posted: 29 Dec 2011 | 0:43:01 UTC - in response to Message 22857.

12/28/2011 7:17:46 PM GPUGRID Requesting new tasks for CPU
12/28/2011 7:17:48 PM GPUGRID Scheduler request completed: got 0 new tasks
12/28/2011 7:17:48 PM GPUGRID Message from server: No tasks sent


I unselected long runs and updated from the Manager, and now it's asking for CPU tasks instead of GPU.

The server software now appears to have a minimum requirement of 238MB for normal task types, and 384MB for long tasks.


Yes, and I have that minimum.

cstaffa
Send message
Joined: 2 Aug 11
Posts: 5
Credit: 737,357
RAC: 0
Level
Gly
Scientific publications
watwatwatwat
Message 22859 - Posted: 29 Dec 2011 | 1:37:13 UTC - in response to Message 22857.

Okay, it finally asked for GPU tasks, with similar results:

12/28/2011 8:13:28 PM GPUGRID Sending scheduler request: To fetch work.
12/28/2011 8:13:28 PM GPUGRID Requesting new tasks for GPU
12/28/2011 8:13:31 PM GPUGRID Scheduler request completed: got 0 new tasks
12/28/2011 8:13:31 PM GPUGRID Message from server: A minimum of 384 MB (preferably 384 MB) of video RAM is needed to process tasks using your computer's NVIDIA GPU
12/28/2011 8:13:31 PM GPUGRID Message from server: No tasks sent
12/28/2011 8:13:31 PM GPUGRID Message from server: No tasks are available for ACEMD2: GPU molecular dynamics
12/28/2011 8:13:31 PM GPUGRID Message from server: No tasks are available for ACEMD beta version


If I am no longer permitted to work on long runs, that sucks, since I have the supposed minimum RAM, and have generally had a better chance of completing a long run without error than a ACEMD or ACEMD2.

MarkJ
Volunteer moderator
Volunteer tester
Send message
Joined: 24 Dec 08
Posts: 738
Credit: 200,909,904
RAC: 0
Level
Leu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 22861 - Posted: 29 Dec 2011 | 7:41:22 UTC
Last modified: 29 Dec 2011 | 7:50:21 UTC

I don't know know if its relevant but your cards have Compute Capability of 1.1. There was talk about restricting tasks to CC 1.3 or better cards. Maybe Toni or GDF can advise if they have implemented that change yet or not.

In the later versions of the BOINC client it indicates how much free memory the cards have at startup, which might help in debugging, however those changes are in the alpha-test version of BOINC.

Also the current release versions of BOINC (6.12.x) honour the "Don't ask for CPU tasks" setting from the server. Seeing as you are running a 6.10.58 client it will try asking for both types of work until the CPU requests get backed-off. Its also quite likely that the 6.10 client (and probably the 6.12 client as well) doesn't pass back the available memory so the server doesn't really know if the card has enough free memory or not. Something for Toni or GDF to look into.
____________
BOINC blog

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 22864 - Posted: 29 Dec 2011 | 13:24:08 UTC - in response to Message 22861.

The server is seeing 383MB rather than the clients reported 384MB:

NVIDIA GeForce 9600 GSO (383MB) driver: 266.58

So it's reasonable to presume that the new server is using the default 384MB cap for long tasks, and possibly short tasks.

So why the difference?
Possibly rounding in the server code; the memory is reported in Bytes, and if the server is reading the left most 3 (or more) numbers, reading from right and skipping the first 6 numbers, it would round down to 383:
For example, 402653184 is exactly 384MB, but if you only read the leftmost 3 numbers, divide by 1024 twice and multiply by 1000 twice (to convert to MB) it's (402/1048576)*1000*1000=383.

AFAIK CC1.1 cards can still get tasks. I think CC1.3+ was going to be introduced around Sept/Oct but was delayed. I think this is still likely to happen soon, within next 3months, given that the GF600 series is due out within that time frame.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

cstaffa
Send message
Joined: 2 Aug 11
Posts: 5
Credit: 737,357
RAC: 0
Level
Gly
Scientific publications
watwatwatwat
Message 22866 - Posted: 29 Dec 2011 | 15:47:37 UTC - in response to Message 22864.

Thank you both for your attention. Thanks to GPUgrid for a good run. I have no plans to upgrade, so I'm shut out.
____________

Viking69
Avatar
Send message
Joined: 11 Aug 07
Posts: 5
Credit: 3,206,127
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwat
Message 23193 - Posted: 31 Jan 2012 | 1:49:23 UTC

OK, I started out with a Nvidia GT 550 w/2 GB DDR3 Ram. It let me join GPUGRID and downloaded a WU. it didn't start processing becaus it said "GPU Missing".
I reloaded the drivers (285.62) twice. And when BOINC started it no longer showed that i had Cuda capable video board.

I replaced the 550 with a GTX 550 ti ( still within my budget) and the system is using 285.62 drivers and It started crunching... WU'Hoooo! Happy.

I retun home today to find that it finished the first WU, downloaded 2 more and I am in the same boat. "GPU Missing"

From my log:
1/30/2012 12:07:34 AM | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz [Family 6 Model 26 Stepping 5]
1/30/2012 12:07:34 AM | | Processor: 256.00 KB cache
1/30/2012 12:07:34 AM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 sse4_2 syscall nx lm vmx tm2 popcnt pbe
1/30/2012 12:07:34 AM | | OS: Microsoft Windows 7: Ultimate x64 Edition, Service Pack 1, (06.01.7601.00)
1/30/2012 12:07:34 AM | | Memory: 4.00 GB physical, 8.00 GB virtual
1/30/2012 12:07:34 AM | | Disk: 244.04 GB total, 192.43 GB free
1/30/2012 12:07:34 AM | | Local time is UTC -8 hours
1/30/2012 12:07:34 AM | | NVIDIA GPU 0: GeForce GTX 550 Ti (driver version 285.62, CUDA version 4.10, compute capability 2.1, 1024MB, 932MB available, 691 GFLOPS peak)
1/30/2012 12:07:34 AM | | OpenCL: NVIDIA GPU 0: GeForce GTX 550 Ti (driver version 285.62, device version OpenCL 1.1 CUDA, 1024MB)
1/30/2012 12:07:34 AM | | NVIDIA GPU is OpenCL-capable


And from this afternnon:
1/30/2012 1:02:58 PM | GPUGRID | Finished download of A1423-TONI_AGGMI1-8-metainp_file
1/30/2012 1:02:58 PM | GPUGRID | Finished download of A1423-TONI_AGGMI1-8-A1423-TONI_AGGMI1-7-100-RND2446_7
1/30/2012 5:30:43 PM | | GPUs have become unusable; disabling tasks


I will reboot to see if this fixes it. CPU-Z did show a graphics card, but now it doesn't. Windows still shows it. Nvidia control panel still says its there.
____________
Finally have a CUDA capable video card.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 23198 - Posted: 31 Jan 2012 | 13:25:31 UTC - in response to Message 23193.

Reset your Bios.
Remove the GPU fully and then reinstall it fully (drivers) - Use driver sweeper to fully remove previous driver installation attempts.
Check your hard disk for corruptions (Windows: Start, Run CMD CHKDSK /F /R).
Confirm that your PSU is powerful enough to support your GPU.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Viking69
Avatar
Send message
Joined: 11 Aug 07
Posts: 5
Credit: 3,206,127
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwat
Message 23212 - Posted: 2 Feb 2012 | 4:19:17 UTC - in response to Message 23198.

Reset your Bios.

I will still do this at some point.
Remove the GPU fully and then reinstall it fully (drivers) - Use driver sweeper to fully remove previous driver installation attempts.

I had done that with the previous unit and it didn't help. It seem to have helped with this GPU though. What and where is driver sweeper?
Check your hard disk for corruptions (Windows: Start, Run CMD CHKDSK /F /R).

This had already been done.
Confirm that your PSU is powerful enough to support your GPU.

I have a Corsair GS 600. So far it has not been taxed in this install.

I am now also running the latest beta 295.51
____________
Finally have a CUDA capable video card.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 23221 - Posted: 2 Feb 2012 | 19:05:39 UTC - in response to Message 23212.
Last modified: 2 Feb 2012 | 19:58:49 UTC

I see you are now running a task, so the driver update worked. If the problem re-occurs reset the Bios, but if not there is no need.

IIRC driver sweeper (or remover/similar) 'fully' removes drivers (including the parts standard uninstalls fail to remove).

Ideally people would use Comodo Programs Manager, which monitors installations and allows full driver uninstalls, but I think you have to install it before you install the driver, so it's best for fresh installations (straight after you install the OS).
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Viking69
Avatar
Send message
Joined: 11 Aug 07
Posts: 5
Credit: 3,206,127
RAC: 0
Level
Ala
Scientific publications
watwatwatwatwat
Message 23584 - Posted: 21 Feb 2012 | 8:08:17 UTC

UPDATE:

I do remote desktop to a lot of my PC's at home. I have noticed that when I do to this box with the NVIDIA board init, the GPU becomes unavailable in BOINC. The board performes fine other wise. To fix, I do a remote reboot.

If I am in as console to my 55"tv all is OK.
____________
Finally have a CUDA capable video card.

MarkJ
Volunteer moderator
Volunteer tester
Send message
Joined: 24 Dec 08
Posts: 738
Credit: 200,909,904
RAC: 0
Level
Leu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 23588 - Posted: 21 Feb 2012 | 11:30:36 UTC - in response to Message 23584.

UPDATE:

I do remote desktop to a lot of my PC's at home. I have noticed that when I do to this box with the NVIDIA board init, the GPU becomes unavailable in BOINC. The board performes fine other wise. To fix, I do a remote reboot.

If I am in as console to my 55"tv all is OK.


This is because most remote desktop software use their own video driver which doesn't handle cuda. There are a few remote desktop programs that work properly, but most don't. To cope BOINC disables the GPU computing when the video disappears and will resume once the device is back (which should be after you break the connection). You shouldn't need to reboot.

If all you need to do is monitor the BOINC clients you can use BOINCtasks or even BOINC manager to connect to a remote machine. BT is good for a number of machines and you can see what they are all up to on a single PC (its windows program but will run under wine)
____________
BOINC blog

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 7,520
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 23589 - Posted: 21 Feb 2012 | 11:52:30 UTC - in response to Message 23584.

To avoid restarting the whole system, you should use other remote control software than Windows' remote desktop, like VNC, LogMeIn, TeamViewer, etc. They "copy" the contents of the screen to the remote instead redirecting the physical layer of the user interface from the host to the remote like RDP does. To achieve a more responsive remote user interface, RDP cuts the connection between the user's desktop (including the applications run by the user) and the host's input/output hardware of the user interface. This connection is restored only when the user logs in from the console again. In the other hand, the other remote control methods require higher bandwith between the host and the remote.

wabr101
Send message
Joined: 3 Feb 12
Posts: 4
Credit: 196,595,724
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 23691 - Posted: 28 Feb 2012 | 18:42:11 UTC

I, too, noticed a few days ago that I had a flat line stat. Checking my logs I had close to 50 WUs error out. I was almost certain that the change I had made was the new nvidia drivers. the 295.xx series.
On a whim I rolled them back to 285.xx series and waited.
Seems OK now.
Card is a GT430--not the fastest, but reliable. Win 7 64 Pro. AMD dual core 5200+.
I should have posted earlier, but wanted to make sure that I had found the problem.

Am wondering now when I build my new 8 core AMD computer how I will get the older drivers since nvidia seems to push the newer ones.

From what i read here, it's not nvidia's fault?
____________

Post to thread

Message boards : Graphics cards (GPUs) : no new WU's, video RAM message

//