GPU Missing

joe areeda
joe areeda
Joined: 13 Dec 10
Posts: 285
Credit: 320378898
RAC: 0
Topic 196132

This was weird.
I'm running Ubuntu 11.10 on an 17 2600K with a GTX 560. nVidia drivers are 280.13 and boinc is 6.12.33 (latest one in the Ubuntu repository). E@H is my only project.

It's been running fine for months averaging 18-22K Credits/day. Well today all the CUDA taks had a message "GPU Missing". Rebooting didn't work, logging off didn't work. Stopping the programs when exiting the manager and nothing came back up.

GPU was shown on the nVidia settings application. The two monitors plugged into that card suggest one of the cats didn't pee on it (I still haven't smelled the case to be sure).

Anyway, the solution was to uninstall and reinstall Boinc. Now it seems to be running fine. I have 3 BPR cuda tasks ready to report so it can't have been down for very long.

This is more a report of something that worked than a request for help. I'll be back if it does it again.

Joe

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 736679216
RAC: 1292344

GPU Missing

Thanks for reporting this. It's weird, but good to know things like this can happen.

HB

joe areeda
joe areeda
Joined: 13 Dec 10
Posts: 285
Credit: 320378898
RAC: 0

This is happening again. I

This is happening again. I just started a new job, (now I get paid to learn about LIGO) and this machine has been pretty much idle except tor E@H tasks. It is my best at home workstation and I haven't been doing much at home. I noticed it by it's dropping average credit.

Rebooting didn't help.

Before I reinstall the Boinc package, which worked last time, can anyone thing of something to check to help identify what's happening?

It seems to be running and reporting CPU tasks just fine, but it doesn't detect the GPU, listing it with out GPU in "my computers" no error messages in the event log.

The nVidia server settings is normal:

Joe

Bikeman (Heinz-Bernd Eggenstein)
Bikeman (Heinz-...
Moderator
Joined: 28 Aug 06
Posts: 3522
Credit: 736679216
RAC: 1292344

Is it a Linux

Is it a Linux box?

Sometimes a kernel update will require that you reinstall any manually installed NVIDIA drivers. Part of the installation is to adapt the driver to the currently installed kernel.

HBE

joe areeda
joe areeda
Joined: 13 Dec 10
Posts: 285
Credit: 320378898
RAC: 0

It is an Ubuntu 11.10 box.

It is an Ubuntu 11.10 box. http://einsteinathome.org/host/4237123

But I decided to use the Ubuntu drivers and not install the very latest from nVidia. I learned the hard way they don't mix very well because they put the same files in different places and I ended p with conflicting versions of the dynamic libraries.

I forgot to check uptime before I rebooted but unless there was a power outage or some kind of automatic reboot it's been up for weeks.

I may change this system to Scientific Linux as the default boot. Frankly I'm not very happy with Ubuntu Oneric. I forced myself to get used to Unity and I like that part but I've had way too many examples of this kind weird behavior.

I'm pretty sure reinstalling Boinc will fix it, at least it did last time and if nobody can think of anything to check first, I'll do that.

Joe

joe areeda
joe areeda
Joined: 13 Dec 10
Posts: 285
Credit: 320378898
RAC: 0

For the record I rebooted

For the record I rebooted this system in Win 7 (http://einsteinathome.org/host/4185301) and Boinc is operating normally.

Also I installed all pending updates in Ubuntu and reboot did not fix it. Also I did not find anything else that didn't work on that system, yet.

I do have a little CUDA learning program I wrote to compare fft times. I'll try that later, when I go back to Ubuntu.

Joe

joe areeda
joe areeda
Joined: 13 Dec 10
Posts: 285
Credit: 320378898
RAC: 0

It's been running Windows for

It's been running Windows for less than 3 hours and has 5 BRP WU's pending.

So it's not hardware. It's going to be working in Windows for a while as I try to figure out why a RAID array works at home but not installed in the production system.

Nathaniel Warner
Nathaniel Warner
Joined: 31 Oct 11
Posts: 2
Credit: 1406787
RAC: 0

I'm having the same problem

I'm having the same problem with my GPU "missing". I have an nVidia GTX 260 running in a Windows Server 2008 x64 box. nVidia drivers are 285.62 and my BIONIC version is 6.12.34 with the event log showing "2/6/2012 8:41:06 PM | | No usable GPUs found".

joe areeda
joe areeda
Joined: 13 Dec 10
Posts: 285
Credit: 320378898
RAC: 0

RE: I'm having the same

Quote:
I'm having the same problem with my GPU "missing". I have an nVidia GTX 260 running in a Windows Server 2008 x64 box. nVidia drivers are 285.62 and my BIONIC version is 6.12.34 with the event log showing "2/6/2012 8:41:06 PM | | No usable GPUs found".


Nathaniel,

Has your GPU ever worked on your Windows box?

Your driver version should be fine. If you post a link to the CPU information on the E@H website someone will tell us if that version of the drivers has any problem with Boinc. The device manager on my Windows 7 box lists its version as 275.33.

The other gotcha is that Boinc must be started from the console. RDP uses a virtual video device that does not have access to the GPU. I've heard that VNC works differently and allows GPU access from Boinc started remotely.

Joe

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2143
Credit: 2962282520
RAC: 692058

RE: RE: I'm having the

Quote:
Quote:
I'm having the same problem with my GPU "missing". I have an nVidia GTX 260 running in a Windows Server 2008 x64 box. nVidia drivers are 285.62 and my BIONIC version is 6.12.34 with the event log showing "2/6/2012 8:41:06 PM | | No usable GPUs found".

Nathaniel,

Has your GPU ever worked on your Windows box?

Your driver version should be fine. If you post a link to the CPU information on the E@H website someone will tell us if that version of the drivers has any problem with Boinc. The device manager on my Windows 7 box lists its version as 275.33.

The other gotcha is that Boinc must be started from the console. RDP uses a virtual video device that does not have access to the GPU. I've heard that VNC works differently and allows GPU access from Boinc started remotely.

Joe


And, to complete the set, for Windows Vista/7/Server 2008 and related OSs BOINC must not be installed in 'Protected Application Execution' or 'service' mode if a GPU is to be used. In order to have access to the CUDA video driver, BOINC must be running under a logged-on user account.

Check the startup messages: if they say 'Running under account [name]', then the GPU should be available. But if they say 'Running as a daemon', you'll have to uninstall BOINC and re-install in user mode.

Nathaniel Warner
Nathaniel Warner
Joined: 31 Oct 11
Posts: 2
Credit: 1406787
RAC: 0

After a little bit of

After a little bit of reconfiguring I got everything working again. I just installed the beta drivers (295.51) for my GTX260, uninstalled BIONIC, rebooted, and then re-installed BIONIC.

Quote:
RDP uses a virtual video device that does not have access to the GPU. I've heard that VNC works differently and allows GPU access from Boinc started remotely.

I am now aware of the effects that RDP has on GPU computing. Earlier today I had remotely logged on to my server multiple times with the aforementioned method, so I'm assuming that's where my problems started. Thank you very much for the help! I'll have to look into VNC and find out if it's a viable alternative.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.