Choose dedicated or partitioned GPU access for Hyper-V workloads

Source facts

Microsoft distinguishes native GPU access on a physical Windows Server from the graphics virtualization needed inside Hyper-V VMs. Discrete Device Assignment dedicates one or more physical GPUs to a VM, which uses the native driver. Starting with Windows Server 2025, GPU partitioning lets multiple VMs receive dedicated portions of a physical GPU rather than each receiving the entire device. Microsoft documentation.

Applicability

Identify the application, guest and host releases, GPU model, driver, and required features. Review the support requirements for the chosen mode, including any guest-platform restrictions, before deciding from density alone.

DSE recommendation

Compare a dedicated-device design with a partitioned design using the actual application requirements. Have the workload and virtualization owners document the required GPU resources and acceptable sharing model. Include maintenance and migration expectations in the design review and retain the supported configuration evidence.

Verification

Run the representative workload in the chosen supported configuration and inspect the GPU resources visible to the guest. Measure the agreed application tasks with the planned number of simultaneous VMs. Record unsupported features or an unmet performance target explicitly before approving the allocation model for wider use.

Official references

Microsoft Learn: Plan for GPU acceleration in Windows Server. Source reviewed September 8, 2026.