omnia use Nvidia galaxy role for driver install

Is your feature request related to a problem? Please describe. Nvidia has a galaxy collection for the nvidia driver that we should investigate: https://galaxy.ansible.com/nvidia/nvidia_driver

Describe the solution you'd like Evaluate if this is a better alternative vs using El Repo

Oct 01 '21 19:10 j0hnL

@j0hnL will have look into it.

Oct 01 '21 19:10 sujit-jadhav

This looks like a far better solution @j0hnL. I'd rather nvidia maintain the codebase than us.

Oct 05 '21 18:10 lwilson

Since we do a lot of other work to make everything function besides just installing the driver, we could use this to replace one of the package installs in this play: https://github.com/dellhpc/omnia/blob/17f47a85aa585f0c70ae46b4bc59f2d0efb69dbb/roles/common/tasks/nvidia.yml#L55-L60

Where {{ nvidia_packages }} is literally 2 just packages: https://github.com/dellhpc/omnia/blob/17f47a85aa585f0c70ae46b4bc59f2d0efb69dbb/roles/common/vars/main.yml#L72-L74

@j0hnL and @sujit-jadhav: With a move to podman will we need to install nvidia-docker2?

Oct 27 '21 14:10 lwilson

@lwilson @sujit-jadhav with podman there we still need the nvidia-container-toolkit

good doc here: https://www.redhat.com/en/blog/how-use-gpus-containers-bare-metal-rhel-8

Oct 27 '21 15:10 j0hnL