GPU setup

Verify your cluster's GPU stack and prepare it for LLMKube. Covers GKE, EKS, and self-managed clusters running the NVIDIA GPU Operator.

What this page will cover

Confirming GPU nodes are present and the device plugin is healthy.
Reading nvidia.com/gpu allocatable counts and matching them to your InferenceService requests.
When to set spec.runtimeClassName: nvidia (clusters where NVIDIA isn't the default runtime).
Validating the install with a tiny GPU smoke-test workload before deploying a real model.