Select a supported Hyper-V GPU partition count

Source facts

Microsoft requires Windows Server 2025 on a GPU-partitioning host; the documented clustered live-migration scenario requires Datacenter. The GPU manufacturer defines the valid partition counts, and administrators select from that supported set rather than inventing a count. The Windows Admin Center procedure requires its GPUs extension at version 2.8.0 or later. Microsoft Learn.

Applicability

Identify the actual GPU model, driver, host edition, and management tooling before dividing capacity among guests. Treat a requested number of workloads as a planning input, not evidence that the hardware supports that many partitions. Confirm the source requirements for the intended topology.

DSE recommendation

Build a placement plan using the counts reported as valid for the installed GPU. Have the workload owner describe the expected demand per guest and choose a supported count for an initial trial. Record which guests will receive partitions and how assignments will be reversed. Keep this plan separate from dedicated-device assignment, which addresses a different allocation method.

Verification

Inspect the available counts and resulting partition configuration on each intended host. Confirm the management view recognizes partitioning before assigning a guest. Exercise the representative workload and record its behavior at the chosen partition size; do not present a successful assignment alone as a performance qualification.

Official references

Microsoft Learn: Partition and assign GPUs to a virtual machine in Hyper-V. Source reviewed September 8, 2026.