EnterpriseNVIDIA

VMware Tanzu Kubernetes Grid

Broadcom / VMware

Kubernetes on vSphere. GPUs are exposed via passthrough (DirectPath I/O) or NVIDIA vGPU, then enabled in-cluster with the GPU Operator.

Bootstrap with VMware Tanzu Kubernetes Grid
Enablement

vGPU / passthrough + NVIDIA GPU Operator

Driver install

GPU Operator in vGPU mode uses a licensed guest driver matched to the host vGPU manager.

Container runtime

containerd with NVIDIA runtime.

Scheduling

Default scheduler; GPU worker node classes in the TKG cluster spec.

Partitioning

vGPU profiles at the hypervisor; MIG on supported cards.

Networking

Antrea (default) or NSX; SR-IOV for higher throughput.

Upgrades

Coordinate guest vGPU driver with the host vGPU manager version on upgrades.

GPU enablement
  1. 1

    Configure the GPU on vSphere

    Set up passthrough or a vGPU profile on the ESXi hosts and the node class.

  2. 2

    Install the GPU Operator in vGPU mode

    Provide the licensed vGPU guest driver and license config.

    bash
    helm install --wait gpu-operator nvidia/gpu-operator \
  -n gpu-operator --create-namespace \
  --set driver.repository=<vgpu-driver-repo>
  3. 3

    Validate

    Confirm the vGPU is visible and licensed inside a pod.

Validate the enablement
verify GPUs
kubectl get nodes -o custom-columns=NAME:.metadata.name,GPU:.status.allocatable.'nvidia\.com/gpu'
kubectl -n gpu-operator get pods
kubectl run cuda-check --rm -it --restart=Never \
  --image=nvidia/cuda:12.4.1-base-ubuntu22.04 \
  --limits=nvidia.com/gpu=1 -- nvidia-smi
Known limitations
  • ·vGPU requires matching host/guest driver versions and licensing.
  • ·Passthrough dedicates a whole GPU to one VM.
Best for

vSphere-standardized enterprises consolidating GPUs with vGPU.