Nvidia driver causes kernel to panic randomly no driver, no CUDA

Accelerated Computing CUDA CUDA Programming and Performance
1

I was getting annoying system freezes on Fedora 12, kernel 2.6.31. The nvidia driver simply refused the start with 2.6.32. By system freeze, I mean the CapsLock buton not working, I mean having to do a hard reset to get things working again, and this happened several times a day.
At first, I believed it might be the kernel modules from VirtualBox. They weren’t the cause. I booted into 2.6.32 and unblacklisted nouveau, and the system has been rock solid for almost a week.

Of course, we have two variables change here: the kernel and the presence of the nvidia kernel module. Either way, this still makes me wonder, WTF!!!

Has anyone had such problems? Is there a way to get the nvidia driver (and thus CUDA) working without having digital sex with the reset button?

2

nouveau is an open source reverse engineered nvidia driver. nvidia has their own that comes with 195… called “nvidia”

3

I’m very well aware of what “nouveau” and “nvidia” are. The nvidia driver causes the kernel panics I mentioned; nouveau is rock-solid, but then I lose 3D, OpenCL and CUDA functionality. Or I use the nvidia driver and take painkillers from pressing the reset button. Not fun.

4

Did you mean that you blacklisted nvidia instead?

I compile my nvidia kernel manually during installation and I have a full kernel build source. There’s a new nvidia driver version out 195.36.24.

Also for me, I cannot have multiple kernels. Every kernel change, I have to reinstall the nvidia drivers.

5

I mean remove “rdblacklist=nouveau” from the kernel command line, and delete xorg.conf. I don’t have to blacklist nvidia, as it simply won’t load if nouveau is present.

lsmod |grep nvidia yields nothing in this configuration.

I have akmod-nvidia-195.36.15 from rpmfusion installed. It automatically builds the kernel module for each new kernel. I’ll try the new driver once it hits rpmfusion. Last time I tried to use the nvidia installer didn’t end too well for me.

6

I understand the situation now, thanks for clarifying.

Ok, and you probably already know…check the dmesg and syslog for any errors. NVRM stuff. Maybe it’s a hardware issue. If windows installs and works, you could stress test the GPU.

7

Also I assume you haven’t overclocked and the memory timings are correct. You can try memtest86 as well. Running cuda and graphics may stress the other components like memory.