[maybe OT?] OS options recommended and Cuda install issues

[charliederr]

if this is innapropriate to bring up here, by all means, send me away (of course having specifics about "where" to go from experts here is a gift i'd happily take with me while I try to avoid being hit in the butt by the door on the way out...)

in searching all the discussions here for the string "fedora", the most recent post (in fact the only one which came up) was from 2023

and that was about a VM install anyhow

i'm looking for information about how to install cuda on bare metal and since the JAX library is the one I am most focused on for the immediate future I just wanted to see what my options might be

if i install fedora as the base OS, will I be able to get cuda 12 and JAX working without too much trouble? or does it make no sense to install anything other than ubuntu on bare metal (my thinking was that I'd be able to use ubuntu docker images or KVM instances on top of fedora if I can get the cuda toolkit working below on fedora and still be able to develop in an ubuntu environment)

pointers towards documentation on any of this, forums that might be helpful, or any other resources eagerly sought

 best,
     ~c

[hawkinsp]

If you install JAX using pip install jax[cuda12], which is what we recommend, then all you need installed in the OS is the NVIDIA driver.

If you use docker, you need to teach docker how to offer the GPU to the container (I think that's nvidia-docker but I don't use docker much.)

Hope that helps?

[charliederr]

Yes, it's all very helpful. Thanks!