Skip to content

NVIDIA GPU Operator

The NVIDIA GPU Operator enables the node GPU for Kubernetes workloads. It installs discovery, validation, metrics, and device-plugin components.

Current implementation

Property Value
Status Active
Operator version v26.3.3
Namespace gpu-operator
Node count 1
Metrics endpoint DCGM exporter on port 9400
GitOps path gitops/infrastructure/controllers

Active components include Node Feature Discovery, GPU Feature Discovery, the NVIDIA device plugin, DCGM exporter, and operator validation.

GitOps ownership

Flux installs the Helm release from gitops/infrastructure/controllers/base/gpu-operator, with the Talos-specific values in gitops/infrastructure/controllers/lab/gpu-operator.

The chart is pinned to v26.3.3. Driver and toolkit installation remain disabled because Talos provides both at the host layer, and hostPaths.driverInstallDir remains /usr/local.

The host preparation is part of the initial Talos installation and is documented in NVIDIA GPU on Talos. Talos owns the NVIDIA kernel modules and container toolkit; the operator owns Kubernetes discovery, validation, metrics, and the device plugin.