Advanced search

Message boards : Server and website : Question on server msg "Fermi Class GPU not supported by cuda2.2"

Author Message
jkwillia2
Send message
Joined: 8 Oct 10
Posts: 4
Credit: 776,782
RAC: 0
Level
Gly
Scientific publications
watwatwat
Message 19211 - Posted: 3 Nov 2010 | 16:27:35 UTC

Getting a interesting message from the server: "Fermi Class GPU not supported by cuda2.2"

This message appears after the message "requesting new tasks for the CPU" or the message "requesting new tasks for the CPU and GPU". This message does not appear after the message "requesting new tasks for the GPU"

Is the message referring to the compute capacity of the GPU?

If that is the case, why would this message appear on my system, since my NVIDIA GTS 450 has a compute capacity of 2.1?

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1919
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 19212 - Posted: 3 Nov 2010 | 16:35:08 UTC - in response to Message 19211.

But you still receive the workunits?

gdf

jkwillia2
Send message
Joined: 8 Oct 10
Posts: 4
Credit: 776,782
RAC: 0
Level
Gly
Scientific publications
watwatwat
Message 19217 - Posted: 3 Nov 2010 | 18:38:29 UTC - in response to Message 19212.

yes, I do. Which begs the question, what is the intention of the server message?

Profile GDF
Volunteer moderator
Project administrator
Project developer
Project tester
Volunteer developer
Volunteer tester
Project scientist
Send message
Joined: 14 Mar 07
Posts: 1919
Credit: 629,356
RAC: 0
Level
Gly
Scientific publications
watwatwatwatwat
Message 19227 - Posted: 4 Nov 2010 | 9:40:20 UTC - in response to Message 19217.

Is anyone else having the same problem?

gdf

Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 918
Credit: 2,223,045,552
RAC: 272,445
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 19230 - Posted: 4 Nov 2010 | 10:09:02 UTC - in response to Message 19227.

Is anyone else having the same problem?

Only if the BOINC work request includes a request for CPU work.

04/11/2010 09:25:18 | GPUGRID | Sending scheduler request: To fetch work.
04/11/2010 09:25:18 | GPUGRID | Requesting new tasks for NVIDIA GPU
04/11/2010 09:25:18 | GPUGRID | [sched_op] CPU work request: 0.00 seconds; 0.00 CPUs
04/11/2010 09:25:18 | GPUGRID | [sched_op] NVIDIA GPU work request: 345.48 seconds; 0.00 GPUs
04/11/2010 09:25:20 | GPUGRID | Scheduler request completed: got 1 new tasks
04/11/2010 09:25:20 | GPUGRID | [sched_op] Server version 611
04/11/2010 09:25:20 | GPUGRID | Project requested delay of 31 seconds
04/11/2010 09:25:20 | GPUGRID | [sched_op] estimated total CPU task duration: 0 seconds
04/11/2010 09:25:20 | GPUGRID | [sched_op] estimated total NVIDIA GPU task duration: 25146 seconds
04/11/2010 09:25:20 | GPUGRID | [sched_op] Deferring communication for 31 sec
04/11/2010 09:25:20 | GPUGRID | [sched_op] Reason: requested by project
04/11/2010 09:25:22 | GPUGRID | Started download of 138-KASHIF_HIVPR_n1_bound_cl_ba2-50-LICENSE
...


04/11/2010 09:37:53 | GPUGRID | Sending scheduler request: To fetch work.
04/11/2010 09:37:53 | GPUGRID | Requesting new tasks for CPU
04/11/2010 09:37:53 | GPUGRID | [sched_op] CPU work request: 1.00 seconds; 0.23 CPUs
04/11/2010 09:37:53 | GPUGRID | [sched_op] NVIDIA GPU work request: 0.00 seconds; 0.00 GPUs
04/11/2010 09:37:55 | GPUGRID | Scheduler request completed: got 0 new tasks
04/11/2010 09:37:55 | GPUGRID | [sched_op] Server version 611
04/11/2010 09:37:55 | GPUGRID | Message from GPUGRID: No work sent
04/11/2010 09:37:55 | GPUGRID | Message from GPUGRID: Fermi-class GPU not supported by cuda2.2
04/11/2010 09:37:55 | GPUGRID | Message from GPUGRID: (reached limit of 2 GPU tasks in progress)
04/11/2010 09:37:55 | GPUGRID | Project requested delay of 31 seconds


(BOINC test version 6.12.4, host 43404, GTX 470)

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 10,094
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 19232 - Posted: 4 Nov 2010 | 10:54:22 UTC - in response to Message 19230.

Have you deselected "Use Central Processing Unit (CPU)" in GPUGrid Preferences?

Richard Haselgrove
Send message
Joined: 11 Jul 09
Posts: 918
Credit: 2,223,045,552
RAC: 272,445
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 19234 - Posted: 4 Nov 2010 | 11:49:32 UTC - in response to Message 19232.

Have you deselected "Use Central Processing Unit (CPU)" in GPUGrid Preferences?

Oh dear. Sorry, but that remark calls for a history lesson.

You'll be aware that there are three entries on the preferences page (there should be four, but that's a BOINC problem we'll come to later). They are:

Use NVIDIA GPU if present (enforced by 6.10+ clients)

Use Graphics Processing Unit (GPU) if available
Use Central Processing Unit (CPU)

The lower pair (believe me, I know about this stuff) control whether or not the GPU/CPU are actually used by work already present on your computer. But they do not control work fetch requests. The client would still request CPU work, even if the result would be a refusal, as now. They are old code, and shouldn't really be necessary any more.

The upper one - which should be paired with a matching "Use CPU (Enforced by version 6.10+)" - actually appeared because of this email, which I sent to the boinc_dev mailing list on 26 September 2009:

Following on from my previous thought, it seems that we aren't paying enough
attention to the scope of some of these settings.

An XML entity has two different scopes:

a scope for the DEFINITION
a scope for the VALUE

Many BOINC preferences are global for both definition and value. They tend
to be stored in prefs.xml (global or override).

Other preferences are project-specific for both definition and value, like
the screen-saver controls Nicolas reminded us of. They live in account.xml
and client_state.xml.

But there's a third class, with a global definition scope but a project
value scope, which seems to me to be under-utilised. In fact, it has only
one member: resource share.

I propose that we examine the recent resource additions, and consider
whether more of them should be moved into the middle scope (global def,
project value). Use/don't use CPU/GPU is an obvious place to start, but
others might be considered: suspend (CPU/GPU) while in use might be another
candidate - different project apps interfere with the desktop to different
degrees, and I might be willing to let GPUGrid, for example, run a CUDA app
while I'm active, but not willing to run an AQUA CUDA app.

These two do control work fetch requests. If I were able to select "Use CPU (Enforced by version 6.10+)", then the client wouldn't issue CPU work requests, and the fact that BOINC has a rogue error message would be conveniently hidden from view.

In fact, GPUGrid already has that control, but it's hidden from view. If GPUGrid were to install a dummy app defined as a CPU app, the control would appear (defaulted, unfortunately, to 'yes'): we'd be able to flip it to 'no': and the error messages would go away, along with a small number of completely unnecessary CPU work requests. That's what AQUA have done with their dummy CUDA app.

But I don't recommend it. BOINC would automatically pick up the presence of the CPU app (not knowing that it was a dummy), and start advertising the project as CPU-capable. That would cause even more confusion than this misleading error message has already done.

Everything I've just said applies equally to the "Use ATI GPU (Enforced by version 6.10+)" control, likewise already present in this project's code, but hidden because no ATI application has been installed (yet).

To my mind, it would be better if these hidden controls could send the 'no' signal, so the work requests could be inhibited for non-existent plan_classes. As and when an ATI or CPU application is installed for real, the controls would become visible (presumably still defaulted to yes) and active under user control - and the automatic updating of the 'supported platforms' list at BOINC projects would work - truthfully - as designed.

Please join me in calling for this (belated) change.

Profile skgiven
Volunteer moderator
Project tester
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 10,094
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 19275 - Posted: 5 Nov 2010 | 11:45:19 UTC - in response to Message 19234.

Thanks Richard, I forgot the lower config was no longer applicable. I'm rather more vaguely aware of long running problems in this area; I seem to have read more speculative posts than factual posts about this issue. You seem to have fuller grasp of the picture than most, if not all, as you point out a method to remedy this. From the end user point of view, it's not fun trying to run Betas and getting unrelated error messages that could have been systemically removed a year ago, one of many such annoyances with GPU crunching. I get the general impression that GPU’s are a pain in the side of Berkeley, and GPU related problems tend to be ignored rather than dealt with. For what it's worth I would belatedly support your motion. Do you think they want to push for Boinc server updates?

I take it this also results in clients asking non-GPU projects for GPU tasks - a Project wide annoyance to everyone with a GPU, and one that delays the collection of WU’s.

Do you think the Additional Work Buffer be moved/re-created for GPU projects only into the third class (global definition scope but with a project value scope) in order to have separate queues for GPU and CPU tasks?
Basically this project needs tasks returned within 2days, so a low cache is a must, but leaves you open to CPU task shortages for other projects (many people still only crunch for one CPU project).

BDDave
Avatar
Send message
Joined: 29 Jul 10
Posts: 6
Credit: 380,428,173
RAC: 80,887
Level
Asp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 19998 - Posted: 21 Dec 2010 | 8:38:00 UTC - in response to Message 19227.

Hi GDF,

Yes, I've been receiving this error message as well exactly as stated above. Unfortunately the tech talk hurts my brain a bit. From what I'm reading, I do not need to do anything, since I'm only fetching GPU work units and not CPU WUs. Just in case please review my info:


Resource share
If you participate in multiple BOINC projects,
this is the proportion of your resources used by GPUGRID 100
Use NVIDIA GPU if present(enforced by 6.10+ clients)yes
Run test applications? no
This helps us develop applications, but may cause jobs to fail on your computer
Is it OK for GPUGRID and your team (if any) to email you?
Emails will be sent from ps3grid@gmail.com; make sure your spam filter accepts this address. yes
Should GPUGRID show your computers on its web site? yes
Default computer location home
Maximum CPU % for graphics 20
Use Graphics Processing Unit (GPU) if available yes
Use Central Processing Unit (CPU) yes

Computer info:
12/21/2010 12:06:34 AM Starting BOINC client version 6.10.58 for windows_x86_64
12/21/2010 12:06:34 AM log flags: file_xfer, sched_ops, task
12/21/2010 12:06:34 AM Libraries: libcurl/7.19.7 OpenSSL/0.9.8l zlib/1.2.3
12/21/2010 12:06:34 AM Data directory: C:\ProgramData\BOINC
12/21/2010 12:06:34 AM Running under account Hal2000
12/21/2010 12:06:34 AM Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU Q8300 @ 2.50GHz [Family 6 Model 23 Stepping 10]
12/21/2010 12:06:34 AM Processor: 2.00 MB cache
12/21/2010 12:06:34 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 nx lm vmx tm2 pbe
12/21/2010 12:06:34 AM OS: Microsoft Windows 7: Home Premium x64 Edition, (06.01.7600.00)
12/21/2010 12:06:34 AM Memory: 4.00 GB physical, 8.00 GB virtual
12/21/2010 12:06:34 AM Disk: 139.73 GB total, 92.03 GB free
12/21/2010 12:06:34 AM Local time is UTC -8 hours
12/21/2010 12:06:34 AM NVIDIA GPU 0: GeForce GTX 470 (driver version 26099, CUDA version 3020, compute capability 2.0, 1248MB, 726 GFLOPS peak)
12/21/2010 12:06:34 AM Einstein@Home URL http://einstein.phys.uwm.edu/; Computer ID 3548128; resource share 100
12/21/2010 12:06:34 AM Milkyway@home URL http://milkyway.cs.rpi.edu/milkyway/; Computer ID 220768; resource share 100
12/21/2010 12:06:34 AM SETI@home URL http://setiathome.berkeley.edu/; Computer ID 5549346; resource share 100
12/21/2010 12:06:34 AM GPUGRID URL http://www.gpugrid.net/; Computer ID 82476; resource share 100
12/21/2010 12:06:34 AM SETI@home General prefs: from SETI@home (last modified 04-Mar-2009 23:03:00)
12/21/2010 12:06:34 AM SETI@home Computer location: home
12/21/2010 12:06:34 AM SETI@home General prefs: no separate prefs for home; using your defaults
12/21/2010 12:06:34 AM Preferences:
12/21/2010 12:06:34 AM max memory usage when active: 2047.56MB
12/21/2010 12:06:34 AM max memory usage when idle: 3685.61MB
12/21/2010 12:06:34 AM max disk usage: 69.87GB
12/21/2010 12:06:34 AM suspend work if non-BOINC CPU load exceeds 25 %
12/21/2010 12:06:34 AM max download rate: 300000 bytes/sec
12/21/2010 12:06:34 AM max upload rate: 300000 bytes/sec
12/21/2010 12:06:34 AM (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)
12/21/2010 12:06:34 AM Not using a proxy
12/21/2010 12:06:34 AM Milkyway@home Restarting task de_separation_16_3s_fix_1_1807426_1292860667_0 using milkyway version 50
12/21/2010 12:06:35 AM Einstein@Home Restarting task h1_1260.65_S5R4__280_S5GC1HFa_0 using einstein_S5GC1HF version 306
12/21/2010 12:06:35 AM Milkyway@home Restarting task de_nbody_model7_3_81942_1292890383_0 using milkyway_nbody version 21
12/21/2010 12:06:37 AM Einstein@Home Restarting task h1_1260.65_S5R4__274_S5GC1HFa_0 using einstein_S5GC1HF version 306
12/21/2010 12:06:37 AM SETI@home Restarting task 11oc10aa.13879.12337.12.10.249_1 using setiathome_enhanced version 610
12/21/2010 12:07:42 AM GPUGRID Sending scheduler request: To fetch work.
12/21/2010 12:07:42 AM GPUGRID Requesting new tasks for CPU
12/21/2010 12:07:44 AM GPUGRID Scheduler request completed: got 0 new tasks
12/21/2010 12:07:44 AM GPUGRID Message from server: No work sent
12/21/2010 12:07:44 AM GPUGRID Message from server: Fermi-class GPU not supported by cuda2.2
12/21/2010 12:09:19 AM GPUGRID Sending scheduler request: To fetch work.
12/21/2010 12:09:19 AM GPUGRID Requesting new tasks for CPU
12/21/2010 12:09:21 AM GPUGRID Scheduler request completed: got 0 new tasks
12/21/2010 12:09:21 AM GPUGRID Message from server: No work sent
12/21/2010 12:09:21 AM GPUGRID Message from server: Fermi-class GPU not supported by cuda2.2


Thank you
Dave
____________

Whiskey
Send message
Joined: 15 Nov 10
Posts: 1
Credit: 16,545,944
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwat
Message 20044 - Posted: 26 Dec 2010 | 19:02:08 UTC

I'm getting the same error message from the server, have been for a while, but now no new work units. GTX460 WIN7-64bit.

cebula
Send message
Joined: 12 Aug 10
Posts: 2
Credit: 153,285,969
RAC: 5,951
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 20243 - Posted: 26 Jan 2011 | 6:41:02 UTC - in response to Message 20044.

Same situation with the error message. It's been going on for quite awhile now. Looking at my tasks here on the site, it seems I completed two about a week ago...but nothing much else. Is there just not much of a demand for Fermi class GPUs right now...? I would find that surprising.

http://www.gpugrid.net/show_host_detail.php?hostid=77714

cebula
Send message
Joined: 12 Aug 10
Posts: 2
Credit: 153,285,969
RAC: 5,951
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 20246 - Posted: 26 Jan 2011 | 14:42:11 UTC - in response to Message 20243.

I fixed it shortly after this post by checking the "Join" tab above. First I downgraded BOINC to the recommended version, then I installed the CUDA3.1 (developer) drivers.

Post to thread

Message boards : Server and website : Question on server msg "Fermi Class GPU not supported by cuda2.2"