- Name: nvidia-docker
- Project Homepage: http://www.nvidia.com/object/docker-container.html
- Why it should be included: allows the use of GPU enabled docker containers
- Is it open source?: yes, but it requires proprietary nvidia drivers
- Latest release tarball: https://github.com/NVIDIA/nvidia-docker/releases/download/v1.0.1/nvidia-docker_1.0.1_amd64.tar.xz
- Who will use it?: particularly useful for machine learning tasks that depend on cuda and cudnn, which isn't available for Solus yet
- Github link: https://github.com/NVIDIA/nvidia-docker
@JoshStrobl. Nvidia docker is a plug-in to docker that manages the interaction between container processes and the Nvidia GPU driver. It handles the driver pass through and mounting of GPU devices in containers. Without it there is no clean way of accessing the GPU inside a container.
It is important to have this functionality for machine learning and any HPC workload. I for one need this support before coming back to Solus.
I found this github repo that got Nvidia docker 2 running on solus. https://github.com/flauted/solus-setup
Something this fundamental should probably be packaged up. Id like to take this on but I don't know enough about solus packaging to take it on.
I forgot to leave a comment here that I published that. Hopefully nvidia-docker.sh helps someone get started. Someone should definitely do it the right way, though. Making bins in an Ubuntu image felt wrong, but building bmake felt hard.