Using NVIDIA vGPU and GPU Pass-through on the Same ESXi Host

Products

• VMware vSphere ESXi

Issue/Introduction

Administrators may want to configure some physical GPUs on a single VMware ESXi host to run NVIDIA vGPU, while assigning other physical GPUs directly to virtual machines using VMDirectPath I/O (GPU passthrough).

A common question is whether the following configuration is supported:

• Some physical GPUs configured for NVIDIA vGPU

• Other physical GPUs configured for PCI passthrough

• All GPUs installed on the same ESXi host

Environment

 

• VMware ESXi

• NVIDIA GPUs supported by NVIDIA vGPU software

• NVIDIA vGPU Manager installed on the ESXi host

• Virtual machines using either:

  • NVIDIA vGPU profiles, or

  • GPU passthrough (VMDirectPath I/O)

 

Resolution

This configuration is supported, provided that each physical GPU is configured for only one mode at a time.

As per the document https://docs.nvidia.com/vgpu/latest/grid-vgpu-user-guide/index.html ,

“GPU pass-through can be used in a server platform alongside NVIDIA vGPU.”

The same document also states:

“A physical GPU can host NVIDIA vGPUs, or can be used for pass-through, but cannot do both at the same time.”

These statements indicate that GPU pass-through and NVIDIA vGPU can coexist on the same server platform, provided that a single physical GPU is not configured for both modes simultaneously.

Therefore, within a single ESXi host, it is supported to configure different GPUs with different modes. For example:

ESXi Host
├ Physical GPU1 → NVIDIA vGPU
├ Physical GPU2 → NVIDIA vGPU
├ Physical GPU3 → NVIDIA vGPU
└ Physical GPU4 → GPU passthrough (VMDirectPath I/O)

 

In this configuration:

• GPUs configured for vGPU are managed by the NVIDIA vGPU Manager.

• GPUs configured for pass-through are assigned directly to a virtual machine using VMDirectPath I/O.

As long as each physical GPU operates in only one mode at a time, such configuration is supported.