Description
Hello,
I have been using a NVIDIA Jetson Orin Nano Developer Kit 8GB for running GPU workloads with Kubernetes, specifically with the K3s distribution, and using the NVIDIA k8s-device-plugin v0.17.1 without any issue (using the Helm chart installation, or with the manifest installation with kubectl).
I just had to install containerd in the Jetson, install and configure the nvidia-container-toolkit version 1.17.7-1 to use the containerd installed directly and CDI, and when deploying the K3s master node, specify the socket of containerd (with the --container-runtime-endpoint flag of K3s server CLI) in order to use our configured containerd as CRI with nvidia runtime, instead of using the default containerd that installs K3s.
I can give some more details of my Jetson setup, such as:
- K8s version: v1.31.5+k3s1
- Jetson Orin Nano DevKit with: JetPack 6.1, L4T 36.4.0
I hope it can be extended the GPU-Operator support for more NVIDIA GPUs, and more specifically for NVIDIA Jetson GPUs in the near future!
Thanks in advance!