nvidia-docker icon indicating copy to clipboard operation
nvidia-docker copied to clipboard

Published 20 hours ago • verified publisher icon NVIDIA

Request: official Fedora support

Open locture opened this issue 3 years ago • 6 comments

I am aware that Fedora was not officially support due to the cgroupv2 and official CUDA support issue.

I recently noticed that Debian 11 had been officially supported, from which I would believe that the cgroupv2 is no longer an issue. On the other hand, official CUDA support has been added for Fedora 35 in the most recent Nvidia CUDA documentation, and packages are now available in the official Nvidia repos.

Is there anything else which stops Fedora from getting official support?

Thanks.

Loc

locture avatar Feb 17 '22 22:02 locture

I did a quick POC to at least get the lowest level component building on fedora35. I'm not sure if / when we will add official support, but at least it appears to build without to much trouble: https://gitlab.com/nvidia/container-toolkit/libnvidia-container/-/merge_requests/143

klueska avatar Feb 18 '22 12:02 klueska

I did a quick POC to at least get the lowest level component building on fedora35. I'm not sure if / when we will add official support, but at least it appears to build without to much trouble: https://gitlab.com/nvidia/container-toolkit/libnvidia-container/-/merge_requests/143

Thanks for your work! I have been currently trying to test it on my Fedora 35 and have yet encountered any issue so far. I would assume that it is ready for use for now and if possible, getting an official support will definitely be something encouraging.

locture avatar Feb 19 '22 02:02 locture

Some updates after 3 months' use:

I have been using the rpm package from the centos repo using the method from https://github.com/NVIDIA/nvidia-docker/issues/553. I did not need to set the cgroupv2 option as it has been supported. It worked without issue so far on my Fedora 35 machine.

However, I do hope that we could get a repo set up for fedora users.

locture avatar May 21 '22 08:05 locture

Thanks for the update @locture. We are still looking at providing Fedora packages, but don't currently have a timeline for it.

elezar avatar May 23 '22 07:05 elezar

Thanks for the update @locture. We are still looking at providing Fedora packages, but don't currently have a timeline for it.

Thank you for your reply @elezar. While I did not look into the code and compiled it by myself, I do wonder how the packages were built against different distros - as to say, is there something highly different among the builds?

I am currently feeling good using the package from the CentOS repo, and I do see a few others working fine with the same installation method. But I do wonder if it is a safe way to go.

locture avatar May 23 '22 09:05 locture

It's probably fine if it is working for you. The only reason we haven't "officially" started supporting it yet is because we don't have testing in place for it. There shouldn't be any danger in using it if it seems to be working though. I would expect it to either work or break spectacularly (which doesn't seem to be the case for you).

klueska avatar May 23 '22 09:05 klueska

With the 1.14.0 release we reworked a lot of our packaging and now have generic RPM packages.

Please see the instructions here: https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html#installing-with-yum-or-dnf and open an issue against https://github.com/NVIDIA/nvidia-container-toolkit if you're still having problems.

elezar avatar Nov 27 '23 11:11 elezar