Is running [K3s inside a Docker container] with GPU support officially supported?

[infikai]

Hi K3s community!

I am looking for some clarification on whether running K3s inside a Docker container (e.g., using docker run rancher/k3s) with NVIDIA GPU passthrough is officially supported, or if there is a recommended path for achieving this.

What I am trying to achieve:
I want to run a containerized K3s cluster (just Agent running inside Docker) and allow my Kubernetes pods to access the host machine's physical GPUs.

What I understand so far:
I know that running K3s directly on a host (bare-metal/VM) with containerd and the nvidia-container-runtime is supported.

However, when K3s itself is encapsulated inside a Docker container, it becomes a "Docker-in-Docker" or "Containerd-in-Docker" scenario, which makes passing through the NVIDIA device plugins and runtimes much more complex. Therefore, I'm looking for some advice here.

Thanks in advance for help!

[brandond]

Why don't you try it and let us know how it works?

[brandond]

Bad bot, go away.