Invalid runtime name: nvidia

Accelerated Computing DGX Spark / GB10 User Forum DGX Spark / GB10
1

Not sure if it related to corrupted docker buildkit ( Corrupt docker buildkit? ) but it does seems that nvidia runtime is missing.

root@gx10-9c8c:/home/andrew# dpkg -l | grep nvidia-container-toolkit
ii  nvidia-container-toolkit                         1.18.0-1                                 arm64        NVIDIA Container toolkit
ii  nvidia-container-toolkit-base                    1.18.0-1                                 arm64        NVIDIA Container Toolkit Base
root@gx10-9c8c:/home/andrew#
exit
andrew@gx10-9c8c:~$ docker run -it --runtime=nvidia --gpus=all nvcr.io/nvidia/cuda:13.0.1-devel-ubuntu24.04 nvidia-smi
docker: Error response from daemon: unknown or invalid runtime name: nvidia

Run ‘docker run --help’ for more information

It is fixed by running nvidia-ctk

nvidia-ctk runtime configure
INFO[0000] Config file does not exist; using empty config
INFO[0000] Wrote updated config to /etc/docker/daemon.json
INFO[0000] It is recommended that docker daemon be restarted.
root@gx10-9c8c:/home/andrew# cat /etc/docker/daemon.json
{
“runtimes”: {
“nvidia”: {
“args”: ,
“path”: “nvidia-container-runtime”
}
}
(failed reverse-i-search)`docer’: cat /etc/^Ccker/daemon.json
root@gx10-9c8c:/home/andrew# service docker restart
root@gx10-9c8c:/home/andrew#
exit
andrew@gx10-9c8c:~$ docker run -it --runtime=nvidia --gpus=all ``nvcr.io/nvidia/cuda:13.0.1-devel-ubuntu24.04`` nvidia-smi
Unable to find image ‘``nvcr.io/nvidia/cuda:13.0.1-devel-ubuntu24.04’`` locally
13.0.1-devel-ubuntu24.04: Pulling from nvidia/cuda
cc43ec4c1381: Pull complete
c7ba38867e8d: Pull complete

1 Like
2

This is not due to an invalid docker buildkit. NVIDIA Container-Toolkit is preinstalled on the DGX Spark but is not fully setup to allow user flexibility.
You can follow the container-toolkit guide for future reference: Installing the NVIDIA Container Toolkit — NVIDIA Container Toolkit

4

Just recieved info that the nvidia runtime is not needed on DGX Spark to have the GPUs visible inside the container, --gpus=all should be sufficient