Running a different run command for docker run
Im trying to use Vulkan inside a pod, the pod is a simple Pytorch + Cuda, nvidia-smi does indeed see the GPU but vulkan not, i saw on a couple of threads that i should run the container with:
the nvidia.json file that should be in the host machine is not copied in the container so vulkan is failing, a lot of issues suggests that the solution might be using --runtime=nvidia as a param of the docker run command
1 Reply
In some cases even tough the driver is correctly loaded one more thing is needed for vulkan to work
xhost +local:docker this should be called to the host machine, from the container is obvisuly uselles
Is there something i can do to achieve this on runpod?