Hacker News Reader

I've been struggling for days getting a rhel8 server with a tesla to work properly with python. All the Cuda samples and graphics driver work, but none of Jax, tf, or pytorch are letting me coerce them into finding the gpu. Maybe this will help?

Oof, i feel your pain. Yes, it sounds like this might help.

Have you tried any of NVIDIAs docker containers?

Here is the pytorch one: https://ngc.nvidia.com/catalog/containers/nvidia:pytorch

They have others for tensorflow etc. I would highly recommend this as the host system only has to have the NVIDIA driver installed with the rest of the dependencies contained in the docker.

Ld load path maybe? I remember this happened to me once a few years back?

How does this compare to https://github.com/openai/triton? Why should I use this and not Triton?

Is it released or will it be on "August 16, 2021" ?

It's released here: https://github.com/NVIDIA/cuda-python

Is it released yet - the installation section looks pretty empty right now.

The post title is really misleading

https://github.com/NVIDIA/cuda-python

That’s great ! Can’t wait to try.

Nvidia Releases CUDA Python