On Thu, 2017-10-05 at 17:07 -0400, m.roth@5-cent.us wrote:

vychytraly . wrote:

...

On Thu, Oct 5, 2017 at 9:51 PM, m.roth@5-cent.us wrote:

...

So, kmod-nvidia installed. Trouble is, I have no tool to test it. And my user might need nvcc, which, of course, is only provided by the NVidia CUDA, which won't install, because it conflicts with kmod-nvidia.

Has *anyone* dealt with this? If so, what was your solution?

Are you installing CUDA from official NVidia repository?

Please don't top post.

Why, is there some other?

I suppose the epel kmod-nvidia might count - it will allow CUDA apps to run but you can't develop with it.

I did try, last week, and went through one failure after another. Actually, my user's testing now with kmod-nvidia. If that doesn't work, I'm back to square one: uninstall everything, then start with CUDA, rather than the proprietary driver. If that doesn't work, uninstall CUDA, then try the most-current proprietary driver.... (The last try wouldn't build, because of the lack of .../include/linux/fence.h.)

I use the nVidia repo for CUDA. It seems to work OK and you don't need to compile anything as far as I can see. Sorry if you already know this, but you can get it from

https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=...

sorry for the long URL - if you want go to

https://developer.nvidia.com/cuda-downloads

and click through the selections. Ultimately you just do a 'yum install cuda' and it installs everything you need for a CUDA development environment including the kernel drivers - and it's kept up to date.

Documentation on it is at

http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#package...

P.

On Fri, 6 Oct 2017, Pete Biggs wrote:

I suppose the epel kmod-nvidia might count - it will allow CUDA apps to run but you can't develop with it.

The ELRepo drivers are just the drivers, not the SDK. That said, my experience is they're packaged much better than the ones nVidia releases as part of the CUDA repo.

The approach I've gone with is to use ELRepo for the drivers, and then use environment modules to provide the CUDA SDK to users. That for me offers little downsides. You easily get to provide multiple releases of the SDK for users, and you get to use the best packaged drivers.

jh