Just wanted to share some into here. I ran into some trouble using the packages from GetDeb: http://www.getdeb.net/release/3624.

I didn't spot the problem until a reboot because I rarely do so.
Here's how it went:
BOINC ran GPUGRID fine immediately after installation (during the same session). After a reboot, however, BOINC tells me that the CUDA device is not found.
I tried the solution suggested by user J to create a symbolic link to libcudart.so and restart BOINC (http://www.gpugrid.net/forum_thread.php?id=592). Creating a symbolic link doesn't fix the problem. Restarting boinc-client, and it sees my video card again.
For the moment, I added the restart command to /etc/rc.local to have it executed automatically on every start up.

If anyone has a better and more permanent solution, please post a reply. Thanks.

Are you using the default Nvidia driver with the distribution, or the one downloaded off the NVIDIA web site?

Mike Doerner

I'm using the beta driver from the repository: 180.11 beta.

I noticed this too tonight, and I hardly ever reboot the machine either. For the record, I'm using 180.22 from http://www.nvidia.com/object/cuda_get.html Thanks for your insight.

That's an old beta. The "newest" release drivers can be found here.....

http://www.nvnews.net/vbulletin/showthread.php?t=122606

Newest beta (that is broken, do not use) is 185.13. I'm using 180.37 (pre-release) right now on SuSE 11.1 64-bit and things are peachy-keen. Make sure you have all the pieces in place on you distro so you can re-compile the kernel, as this driver makes it's own modules. Good luck.

Mike Doerner

This must be "Linux Re-Boot" day. I had to shut down my Ubuntu 8.10 box today also....for some wiring changes..But I am still on the old drivers (177.80 and 6.5.4), so no problems.

Sounds like I might have to try a newer 64 bit driver (180.37..maybe..), not the Beta though...;)
____________


Consciousness: That annoying time between naps......

Experience is a wonderful thing: it enables you to recognize a mistake every time you repeat it.

Latest release (non-beta) is 180.29. FYI.

Mike Doerner

I had the same problem with the getdeb boinc package. I dont think it does CUDA work for some reason. I upgraded manually to 6.6.something and my GTX 280 was found. Ill post here what cruncherpete posted on my team forum on how to install the standard boinc over the package and get it to work normally.

I have a strange problem. Yesterday I've installed nvidia drivers from their site, version 180.22 on my computer with GPU 8600GT. Afterwards I installed boinc 6.45 from getdeb, copied libcudart.so to /var/lib/boinc-client and everything worked fine, boinc could find cuda.
In the meantime I've changed graphic card to 9800GTX+, Ubuntu immediately found the new card, but now boinc could not see cuda. Also I've noticed some problems with 3d games, they didn't run smoothly. I've reinstalled over the nvidia drivers and cuda, but no change.
I put the old card on another computer that already run the new nvidia drivers, and the same problem after the change of card, so I wouldn't say that it's hardware problem.

When I go to NVIDIA X Server Settings everything seems OK, I can see the card, temperature, I can set the options and so on, but for some reason it's not working as it should.

Did you try re-installing the driver as root outside of X-Windows? BTW, 180.29 is the current release for Linux.

Mike Doerner

Did you try re-installing the driver as root outside of X-Windows? BTW, 180.29 is the current release for Linux.

Mike Doerner

I just installed an 8800GT into a headless cruncher that runs Dotsch/UX (Ubuntu v8.10) 64b. I ran the restricted driver thing and it says that version 180 is active. However after that I rebooted and..

1) Right after "Running DKMS automatic installation....kernel v2.6...."
Nvidia 180.11 [fail]
2) After sudo /etc/init.d/gdm start it get's a panel of nvidia errors (possibly xorg) the first of which is "module not loaded "type1" (not found)"
3) Ubuntu low res error box. Telling it to go on and run in low res results in what appears to be it's normal (1024x768 I think) desktop.
4) BOINC (v6.4.5) reports No CUDA devices found.

For tonight disabled the v180 driver, did a sudo dpkg-reconfigure xserver-xorg, turned the onboard video back in in BIOS, pulled the 8800 out and rebooted it so it can crunch on the CPU till tomorrow when I hope to tackle this again.

Any suggestions? Is there a noob/idiot proof Ubuntu step by step anyplace?

Thanx in advance.
____________
- da shu @ HeliOS,
"A child's exposure to technology should never be predicated on an ability to afford it."

Tonight I took my HD4770 out of my desktop and put another 8800GT into it. The restrict hardware driver pop-up came up, I told to install the v180 driver and all seemed good. The nvidia xorg reconfig thing must have run because after a reboot it had the nvidia stuff in it. I added the 'coolbits' '1' to it.

And restarted BOINC... nothing new there... Oh yea, I was still running 6.2.14 on this machine. I downloaded and installed v6.4.5 from getdeb and then got the missing cudart error. Found a post on that and created a link to the missing library in /var/lib/boinc-client.

Restarted again and CUDA recognized! Attached to GPUgrid and am crunching my 1st WU as I type... :-D PARTY!!!!

Now to get that other 8800GT working in the cruncher.

Anybody have any idea what sort of GPU core temp I should be seeing on an 8800GT doing GPUgrid work? It's very slighly OC'd at core=654, mem=957

I have a solution to the no CUDA devices found:

as root (su command) issue chmod a+rw /dev/nvidia*

stop the boinc client before issuing the command

sudo /etc/init.d/boinc-client stop (dont forget to restart)

This seems to work I dont know why.. the command has to be issued each time you reboot..