In reply to @grw00:matrix.org

hey all, has anyone had success using cuda libraries inside a docker container built with nix? i don't mean running a cuda container on nixos host but the opposite, running a nix container containing cuda program on another host
i build a container with nix and pytorch etc and run it on runpod, it doesnt see nvidia drivers/device though, i guess i am missing something. currently i have:

        dockerImages.default = pkgs.dockerTools.streamLayeredImage {
          name = "ghcr.io/my-image";
          tag = "latest";

          contents = [
            pkgs.bash
            pkgs.uutils-coreutils-noprefix
            pkgs.cacert
            pkgs.libnvidia-container

            pythonEnv
          ];

          config = {
            Cmd = [ "${pkgs.bash}/bin/bash" ];
            Env = [
              "CUDA_PATH=${pkgs.cudatoolkit}"
              "LD_LIBRARY_PATH=${pkgs.linuxPackages_5_4.nvidia_x11}/lib"
            ];
          };
        };

In reply to @ss:someonex.net
Hard coding linuxPackages in the image is a bad idea. With cuda you normally don't want drivers in the image, you want the host's drivers mounted in the containet

In reply to @grw00:matrix.org
not sure what CDI is, i understand i need the /run/opengl-driver but i'm not sure how to achieve that in docker container

In reply to @grw00:matrix.org
ah kk, got it. i'm specifically trying to use this on runpod.io, i don't think they offer this as a possibility. it seems like the images they offer all have cuda installed in image