Configuring NVIDIA Virtual GPU (vGPU) in a Linux VM on Lenovo ThinkSystem Servers

Virtual GPU (vGPU) is the method of virtualizing a GPU adapter installed in a server so that the physical GPU can be shared by multiple virtual machines running on that server. This method assigns the virtual devices to multiple guest operating systems, and the guest OSes share the performance of the single physical GPU.

This paper provides guidance on enabling vGPU and making it available to VM running in a Kernel Virtual Machine (KVM)-based OS. The paper is for Linux administrators wishing to use a vGPU in a ThinkSystem™ server to pass through to a VM. For our evaluation, we implement Red Hat Enterprise Linux 8.5 both as the host OS and guest OS, and show how to virtualize the resources of an NVIDIA T4 GPU.

Introduction
Setup NVIDIA vGPU
Create NVIDIA vGPUs
Installing the vGPU driver on the Linux guest OS
Deleting a vGPU on Linux with KVM
Prerequisites for using NVIDIA vGPU on the NVIDIA Ampere architecture
Resources

To view the document, click the Download PDF button.

Related product families

Product families related to this document are the following:

• GPU adapters
• Red Hat Enterprise Linux
• SUSE Linux Enterprise Server