I’ve just got my Tesla c1060 to work alongside an ATI card under Ubuntu 8.04. The card passes the deviceQuery and Bandwidth tests, and the nbody benchmark works (I haven’t tried any others yet).
My problem is that running nvidia-settings causes the screen to go black, and then I return to the login screen. Has anybody else had this problem? The second problem I refer to in the title may or may not be related - the screen will go black at random for a couple of seconds, and then my desktop will reappear. It happened twice while I was typing this.
I’ve attached my xorg.conf. The section about the Tesla outputting to a ‘virtual’ screen is the thing that makes the card work. If I take it away, nvidia-settings loads up, but says that the nvidia driver isn’t working (and its true - in this case, the deviceQuery doesn’t see anything).
Stop doing what you’re trying to do, I think–you’re just going to hit weird problems and have a watchdog timer on the C1060. Instead, try initializing the NVIDIA driver with the script in the 2.1 release notes.
Thanks for the responses. Netllama: I can’t find the nvidia-bug-report.log anywhere on my computer. The script file to generate the log is in /usr/bin, but the log itself isn’t anywhere to be found.
I tried adding the startup script from the release notes, but it didn’t do anything. I tried this with the xorg.conf that I sent, and then with ‘load “nvidia”’ removed, but the behavior in both cases is exactly the same. If I use the startup script but remove the virtual screen references from xorg.conf, then the card doesn’t work, as before.
I’m not sure what the script does. The appropriate /dev/nvidia* files were being generated without the script (I checked).
With regard to the ‘watchdog’ comment, I’m pretty sure that the benchmark ‘nbody’ ran for more than 5 seconds, so it doesn’t seem like there is a watchdog present.
…I tried running the nvidia-bug-report script manually, and when I did so I got the same black screen/back to login behavior as I do with nvidia-settings. When I log back in, there is no bug report log file to be found.
A friend of mine suggested that the problem is that nvidia-settings is trying to display output on the virtual screen associated with the Tesla GPU, hence the X crash. I tried using nvidia-settings --display=0.0 --ctrl-display=1.0, but it tells me that it doesn’t recognize display 1.0. Clearly I’m not specifying the virtual display correctly. Does anyone have any idea how to do this? Any comments on this approach?
The nvidia driver does appear in my xorg.conf. I load it as a module, then I specify it as the driver for the Tesla device. I assumed that this is what created the correct entries.
What I’m saying is do not load it in xorg.conf. Nothing about the NV card should be in your xorg.conf. If the xorg.conf is entirely devoid of NVIDIA-related information and then you run the 2.1 script, everything should be okay. Somebody else tried that once and claimed it worked okay, so that’s what I’m assuming is the case here.
I removed all traces of the nvidia and tesla card from xorg.conf, and sure enough the tesla card works. (That is, it passes the deviceQuery and other tests.)
The nvidia-settings doesn’t crash X any more, but now tells me that the nvidia driver isn’t working, and that I should run nvidia-xconfig. This just replaces the “radeon” driver in the xorg.conf device entry for my ATI display card with “nvidia”. X fails to load with this setting (unsurprisingly). I searched around the forums a little, but I couldn’t find a solution to this problem.