vGPU Enabled ESXi hosts failing to perform DRS migration for VMs on busy hosts

Products

• VMware vCenter Server

Issue/Introduction

Symptoms:

• Your hosts are reaching high memory or CPU load in DRS clusters with vGPU-enabled VMs.
• vGPU-enabled VMs fail to migrate despite a fully automated DRS cluster configuration with low DRS scores.

Environment

ESXi 7.x
ESXi 8.0 U1

Cause

DRS is unable to evaluate stun time for these VMs as vGPU DRS load balancing is not supported until ESXi 8.0 U2.

"Starting with vSphere 8.0 U2, DRS can automatically migrate VMs with vGPUs."

Resolution

Permanent solution:

Upgrade ESXi hosts to a version later than 8.0 U2.

Workaround:

In each VM's configuration options, change the vMotion Stun Time Limit option (VM Options > Advanced) to a value of 101.


Please note: this workaround may not work perfectly, and if it does not, the only other solution is to upgrade your hosts. DRS is NOT supported for vGPU-enabled VMs prior to ESXi 8.0 U2.