Expose ability to run containers with access to GPUs

Question

Expose ability to run containers with access to GPUs

luisquintanilla opened this issue 10 months ago · 3 comments

One of the prerequisites for running AI container workloads on NVIDIA GPUs is the NVIDIA Container Toolkit.

Once that's installed, when you run the container, you need to pass the --gpus flag.

Right now there doesn't appear to be a way to configure a container resource in the AppHost project such that it can be started with GPU support, e.g. on Docker for Windows, the container must be run like docker run -d --gpus=all ....

Answer 1 · 2024-03-11T18:16:54.000Z

This is tracked by https://github.com/microsoft/usvc/issues/141

Answer 2 · 2024-03-27T16:19:59.000Z

@davidfowl. Looks like this is going to need to be pushed to Preview 6? Any objections?

Answer 3 · 2024-03-27T17:47:09.000Z

Moving to P6. I don't see us getting the PR through on the app model side for P5.