Feature Overview

Enable live migration of instances with virtual Graphics Processing Units (vGPU) for the Red Hat OpenStack Platform. 

Goals

This epic aims to introduce enablement for live migration of OpenStack instances with vGPUs allowing users to seamlessly leverage GPU acceleration for their AI workloads. 

Requirements 

• qemu-kvm-8.0.0/libvirt-9.5.0 shipped with RHEL 9.3 can support vGPU live migration with a single vGPU device
• qemu-kvm/libvirt in RHEL 9.4 will support multiple vGPUs

Use Cases

• Use Case 1: As an administrator, I can ensure the uninterrupted availability of vGPU-accelerated instances during hardware maintenance or unexpected hardware component failures by live migrating instances with vGPUs
• Use Case 2: As a user, I can move my workloads running on instances accelerated by vGPUs without causing data loss.