Hacker News Reader

Comment by Patrick_Devine | Hacker News Reader

I think leveraging GPUs inside of Docker containers is an absolutely fantastic idea, but forking/wrapping Docker to make it happen is just silly. Were there no attempts to talk to the core team to make this happen?

(Disclaimer: I work at Docker, but not on the core team)

View on HN

We are only wrapping the Docker CLI, not forking the full code (that would be insane). The wrapper is provided for convenience since it should be enough for most users. If you know what you're doing, you don't need to rely on the wrapper, we explain how you can do that on our wiki: https://github.com/NVIDIA/nvidia-docker/wiki/Internals

Yes, we have been discussing with a few people at Docker, I hope attending DockerCon and discussing with the team will help us move forward.

This is a fantastic use case, we (Docker) would love to help in any way we can to make this usable by all Docker users.

Let us know if there's anything more we can to help.

https://github.com/NVIDIA/nvidia-docker/wiki/Why%20NVIDIA%20...

> However, this solution was brittle: the version of the host driver had to exactly match driver version installed in the container.

I'm curious why volume mounting is not an option here. Seems like you'd just ship driverless containers, forcing the user to mount the host drivers before allowing the container to run.

That's exactly what we do, the image is indeed driverless and we mount the host driver files as a volume (provided by our volume plugin) when the container is launched. This way, you can launch a CUDA 7.5 container on any machine with driver version 352+

We provide more details here: https://github.com/NVIDIA/nvidia-docker/wiki/NVIDIA-driver Hope that helps!

Reply on news.ycombinator.com