Issue: GPU Operator Fails on Jetson Orin (ARM64) — Needed for Kai Scheduler

[Ashwinraj2000]

I'm running Kubernetes (K3s v1.32.4) on a Jetson Orin NX 16GB and trying to use the NVIDIA GPU Operator v25.3.0 for using Kai Scheduler, which relies on DCGM-based GPU telemetry.

Despite customizing the Helm chart with:

nodeSelector.kubernetes.io/arch=arm64

the operator fails to initialize the ClusterPolicy controller due to API server communication errors:

error: failed to get restmapping: failed to get server groups: Get "https://10.43.0.1:443/api": net/http: TLS handshake timeout

This happens consistently even though the cluster is healthy (kubectl get nodes reports Ready), and the node is running JetPack 5.1.3 with CUDA and drivers preinstalled.

Environment

• Jetson Orin NX 16GB

• JetPack 5.1.3 / L4T R35.5.0

• K3s v1.32.4+k3s1

• Ubuntu 20.04 (aarch64)

• Helm 3.13

• Kai Scheduler