Turkey (Turkish)
Germany (German)
Worldwide (English)
Microsoft GPU-P documentation targets Windows Server 2025 and later.
Halkali merkez neighborhood fatih st ozgur apt no 46 , Kucukcekmece , Istanbul , 34303 , TR
GPU-P differs from classic DDA passthrough by partitioning physical GPU resources across VMs. A PowerShell command alone is not enough: supported server-class hardware, the correct driver/vGPU stack and Windows Server version are prerequisites.
Do not assume a desktop RTX card is production-supported merely because GPU-P cmdlets exist. Microsoft documentation explicitly separates server-class support, driver and licensing requirements.
Windows Server 2025 GPU-P attaches a partition of a supported GPU to a VM; current Microsoft guidance notes one GPU partition per VM. Live migration adds GPU/vGPU software and version requirements.
Windows Server 2025 GPU-P attaches a partition of a supported GPU to a VM; current Microsoft guidance notes one GPU partition per VM. Live migration adds GPU/vGPU software and version requirements.
Partition a physical GPU across Hyper-V VMs with Windows Server 2025 GPU-P. Validate hardware, PowerShell state, drivers, VM configuration and live-migration requirements.
Microsoft GPU-P documentation targets Windows Server 2025 and later.
Current troubleshooting guidance notes one GPU partition per VM.
Windows Server 2025 includes supported GPU-P live-migration scenarios.
Microsoft troubleshooting guidance cites NVIDIA vGPU v18+ for live migration.
GPU-P partitions GPU resources; DDA directly assigns a device to one VM. Choose based on isolation, density and vendor support.
| Model | Sharing | Typical use |
|---|---|---|
| GPU-P | Yes, partitions | Multiple VMs per GPU |
| DDA | No, device passthrough | One device → one VM |
Validate Hyper-V, GPU drivers, IOMMU/firmware and partitionable GPU state first. Do not force past unsupported status.
Get-WindowsFeature Hyper-VGet-VMHostPartitionableGpuGet-PnpDevice -Class Display | Format-Table Status,FriendlyName,InstanceId -AutoGet-CimInstance Win32_VideoController | Select-Object Name,DriverVersion,AdapterRAMCheck IOMMU/VT-d/AMD-Vi, Above 4G Decoding and vendor-recommended firmware settings for the server model.
Add the GPU partition adapter while the VM is off and tune min/optimal/max resources only within supported guidance.
$vm='AI-VM01'Stop-VM -Name $vm -ForceAdd-VMGpuPartitionAdapter -VMName $vmGet-VMGpuPartitionAdapter -VMName $vm | Format-List *Start-VM -Name $vmSeeing a device in Device Manager is insufficient. Run CUDA/DirectML/vendor APIs and a real inference or rendering workload.
nvidia-smidxdiag /t C:\Temp\dxdiag.txtGet-PnpDevice -Class DisplayGPU-P live migration can behave differently from standard VM migration due to GPU state transfer and vendor-stack requirements. Measure with a staging VM first.
Splitting one GPU across four VMs does not guarantee each receives a fixed quarter of performance. Workload type, scheduling and vendor profiles matter.
| Metric | Measurement |
|---|---|
| VRAM | Peak + headroom |
| GPU utilization | p50 / p95 |
| Latency | Single VM vs concurrent VMs |
| Migration | Duration + workload impact |
Plan a physical GPU configuration on Eka Sunucu around target VM count, VRAM, workload and Hyper-V topology.
Primary documentation and technical references used by this guide.
Continue with related infrastructure and implementation guides.
Windows Server GPU-P
This guide targets Microsoft’s Windows Server 2025 GPU partitioning support model. Do not assume the same support for client OS or desktop-class hardware.
Current Microsoft troubleshooting guidance states support for one GPU partition per VM.
Windows Server 2025 supports scenarios for it, subject to GPU/vGPU stack, version and hardware requirements.
Cmdlet availability is not a support statement. Check Microsoft/vendor support matrices and licensing before production design.