Research GPU use in Docker and Podman on Windows

[davidpanderson]

Describe the bug

Figure out how to run a CUDA program in Docker on WSL and Windows:

• What needs to be in the Dockerfile?
• What's the Docker run command?
• Does anything special (driver, CUDA toolkit) need to be installed in the WSL image, or on the host (Win)?
• All the above for Podman

Possibly useful:
https://docs.nvidia.com/ai-enterprise/deployment/vmware/latest/docker.html
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html#installing-the-nvidia-container-toolkit
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/cdi-support.html

And then:
how to access CUDA on Linux and Mac?
how to access AMD GPUs on various platforms?
how to access Apple GPUs on Mac?

[aptalca]

For nvidia, container toolkit and the nvidia drivers need to be installed on a linux host. On Windows, docker desktop includes the toolkit.

The container needs to be created with either runtime=nvidia or gpus=all. Then you need the opencl-icd bits in the container.

The linuxserver docker image works with cuda: https://github.com/linuxserver/docker-boinc