NVIDIA vGPU

NVIDIA vGPU is virtual GPU software that lets multiple virtual machines (VMs) share one physical NVIDIA GPU, each with a fixed slice of its memory. The NVIDIA Virtual GPU Manager runs on the hypervisor, and a guest driver runs in each VM. A vGPU is either time-sliced, with VMs taking turns on the GPU's engines, or backed by a Multi-Instance GPU (MIG) partition the VM owns.

On the RTX PRO 6000 Blackwell Server Edition, time-sliced profile DC-96Q gives one VM all 96 GB, and DC-3Q gives 32 VMs 3 GB each. Q-series profiles need an RTX Virtual Workstation (vWS) license, B-series need Virtual PC (vPC) or vWS, and A-series need Virtual Applications (vApps). Each VM checks out its license over the network from the NVIDIA License System. Performance degrades if a VM fails to obtain one within 20 minutes. vGPU for Compute is licensed only through NVIDIA AI Enterprise. Passthrough for professional 3D graphics on these GPUs also takes a vWS license. Supported hypervisors include VMware vSphere ESXi, Nutanix AHV, and Red Hat Enterprise Linux with KVM.

Choose vGPU when many users or services each need a slice of one data center GPU. Choose GPU passthrough when one VM needs the whole card.

Sources

Last verified October 7, 2026.