Installing NVIDIA vGPU on Kubernetes Cluster with RTX 6000 Ada Without Hypervisor

Virtual GPU Forums More vGPU Forums NVIDIA Virtual GPU Technology
1

Installing NVIDIA vGPU on Kubernetes Cluster with RTX 6000 Ada Without Hypervisor

Hello NVIDIA Community,

I have set up a Kubernetes cluster using kubeadm, consisting of:

  • 1 Master Node
  • 2 Worker Nodes

All nodes are running Linux as the default operating system, and the cluster is already configured and operational.

I am using RTX 6000 Ada GPUs across the worker nodes and want to install NVIDIA vGPU software on this setup without using a hypervisor. My requirements are:

  1. Utilize vGPU capabilities directly on the Linux OS across the worker nodes with RTX 6000 Ada GPUs.
  2. Efficiently allocate GPU resources to pods running on the Kubernetes cluster.
  3. Maintain compatibility with the existing cluster configuration and avoid reformatting nodes for a hypervisor setup.

Is it possible to achieve this configuration with the current NVIDIA vGPU software?
If so, could you provide guidance on:

  • The necessary steps and prerequisites for installation.
  • Any potential limitations or licensing requirements for this deployment method.
  • Best practices for integrating vGPU with the NVIDIA device plugin in Kubernetes.

Your assistance would be greatly appreciated, as I aim to optimize my cluster’s GPU utilization while maintaining the current Linux environment.

Thank you in advance for your support!

Best regards,
Venkat Manish