Windows Server 2025 Hyper-V GPU Partitioning (GPU-P) & Discrete Device Assignment (DDA) for High-Performance RDP Workstations: Complete PowerShell & Driver Provisioning Runbook

Virtualizing high-performance graphic workloads on Microsoft Windows Server has undergone a radical architectural transformation. In legacy deployments, remote desktop sessions relied on RemoteFX vGPU, a virtualized software-rasterization layer that Microsoft officially deprecated and removed due to severe architectural vulnerabilities and lack of modern DirectX 12, Vulkan, and CUDA compute support^[1].

When running Remote Desktop Protocol (RDP) sessions on unaccelerated Hyper-V virtual machines, the operating system defaults to the Microsoft Basic Render Driver (WARP). This forces the host CPU to perform software emulation for desktop composition, video decoding, and 3D rendering. The result is sluggish UI interaction, high CPU spikes, and severe frame drops for remote developers, 3D modelers, and multi-monitor workstations.

Windows Server 2025 provides two enterprise-grade hardware acceleration mechanisms to eliminate this bottleneck: Discrete Device Assignment (DDA) for exclusive PCIe passthrough^[2], and GPU Partitioning (GPU-P) for shared hardware slicing across multiple concurrent virtual machines^[3]. Combined with Remote Desktop Session Host (RDSH) Group Policies enforcing H.264/AVC 444 profile streaming^[4], administrators can achieve true 60 frames-per-second (FPS) interactive performance over standard RDP connections.

---

Architectural Comparison: DDA vs. GPU-P

Choosing between Discrete Device Assignment and GPU Partitioning depends on your workload tenancy requirements, resource isolation constraints, and GPU hardware vendor architecture:

Architectural Metric	Discrete Device Assignment (DDA)	GPU Partitioning (GPU-P)
Passthrough Granularity	Dedicated physical PCIe device assignment (1:1 VM to GPU ratio)[2].	Fractional GPU slicing via paravirtualized WDDM drivers (1:N ratio)[3].
Compute & API Support	Native Bare-Metal API access: Full DirectX 12 Ultimate, Vulkan 1.3, CUDA, OpenCL, OptiX[1].	Direct3D 11/12 and Vulkan rendering; compute (CUDA) varies by vendor and driver tier[3].
Hardware Requirements	CPU VT-d / AMD-Vi with Access Control Services (ACS) & SR-IOV BIOS support[2].	WDDMv2.5+ compatible enterprise or consumer GPU with SR-IOV or partitioning support[3].
Memory Isolation	Dedicated Physical VRAM allocation directly mapped into VM MMIO space[1].	Dynamically shared or bounded VRAM slices partitioned across guest instances[3].
Primary Use Cases	Heavy 3D CAD, Unreal Engine rendering, AI model inference, CUDA compilation, dedicated engineering workstations.	Multi-user development pools, remote office productivity, lightweight graphic acceleration, multi-tenant RDS farms.

For production enterprise workloads requiring deterministic performance, uncompromised CUDA execution, and strict physical isolation, deploying Dedicated Servers configured with discrete enterprise GPUs guarantees full hardware access without hypervisor resource contention.

---

Host Prerequisites and BIOS Verification

Before provisioning DDA or GPU-P on Windows Server 2025, verify that your host motherboard and processor meet the required virtualization standards:

1. SR-IOV (Single Root I/O Virtualization): Must be enabled in the BIOS/UEFI under PCIe Configuration.
2. IOMMU (Intel VT-d or AMD IOMMU / AMD-Vi): Required for DMA remapping and memory translation protection.
3. PCIe Access Control Services (ACS): Ensures peer-to-peer PCIe transactions do not bypass the IOMMU between different PCIe root ports^[2].

Verify host DDA support using the official Microsoft Hyper-V PowerShell validation script:

# Query all graphics devices on the physical host
Get-PnpDevice -Class Display | Select-Object FriendlyName, InstanceId, Status

# Confirm Hyper-V role and SR-IOV support
Get-VMHost | Select-Object LogicalProcessorCount, MemoryCapacity, IovSupport

---

Provisioning Discrete Device Assignment (DDA): Step-by-Step

In a DDA deployment, the physical GPU is dismounted from the host operating system kernel and assigned exclusively to the target Hyper-V virtual machine^[1][2].

Step 1: Identify Device Location Path

Locate the exact PCIe location path of the target graphics adapter using PowerShell:

$GpuDevice = Get-PnpDevice -Class Display | Where-Object { $_.FriendlyName -like "*NVIDIA*" -or $_.FriendlyName -like "*AMD*" -or $_.FriendlyName -like "*Intel*" } | Select-Object -First 1
$LocationPath = (Get-PnpDeviceProperty -InstanceId $GpuDevice.InstanceId -KeyName "DEVPKEY_Device_LocationPaths").Data[0]

Write-Host "Target GPU Instance ID: $($GpuDevice.InstanceId)"
Write-Host "PCIe Location Path: $LocationPath"

Step 2: Disable and Dismount GPU from Host

Disable the GPU on the parent host and dismount it from the host PCIe bus:

# Disable device on host OS
Disable-PnpDevice -InstanceId $GpuDevice.InstanceId -Confirm:$false

# Dismount device from host hypervisor
Dismount-VMHostAssignableDevice -LocationPath $LocationPath -Force

Step 3: Configure Target VM Memory and MMIO Space

GPUs feature massive onboard VRAM requiring high memory-mapped I/O (MMIO) allocation in the VM configuration^[1]. For a 16GB or 24GB GPU, allocate sufficient High MMIO space:

$VMName = "WinServer2025-GPU-Workstation"

# Stop VM before modifying assignable devices
Stop-VM -Name $VMName -Force

# Set automatic stop action to TurnOff (required for DDA)
Set-VM -Name $VMName -AutomaticStopAction TurnOff

# Enable Guest Controlled Cache Types for optimal graphics caching
Set-VM -VMName $VMName -GuestControlledCacheTypes $true

# Allocate Low MMIO (128MB) and High MMIO (32GB for large VRAM buffers)
Set-VM -VMName $VMName -LowMemoryMappedIoSpace 134217728
Set-VM -VMName $VMName -HighMemoryMappedIoSpace 34359738368

Step 4: Assign GPU to VM and Boot

Assign the physical device to the virtual machine and start the VM:

# Assign PCIe device to VM
Add-VMAssignableDevice -VMName $VMName -LocationPath $LocationPath

# Start VM
Start-VM -Name $VMName

Once the virtual machine boots, install the vendor’s enterprise production drivers (e.g., NVIDIA RTX Enterprise or AMD Software: PRO Edition). The device will appear directly in Device Manager as a native physical graphics card.

---

Provisioning GPU Partitioning (GPU-P): Multi-Tenant Slicing

When multiple virtual machines or remote desktop users must share a single high-capacity physical GPU, GPU Partitioning (GPU-P) provides paravirtualized graphics slicing without requiring expensive dedicated vGPU licensing tiers^[3].

Step 1: Assign Partition Adapter via PowerShell

Execute the following commands on the Windows Server 2025 host:

$VMName = "WinServer2025-RDSH-Pool01"

# Stop target VM
Stop-VM -Name $VMName -Force

# Assign GPU Partition Adapter
Add-VMGpuPartitionAdapter -VMName $VMName

# Configure partition resource bounds (Optimal dynamic ratio)
Set-VMGpuPartitionAdapter -VMName $VMName `
    -MinPartitionVRAM 1073741824 `
    -MaxPartitionVRAM 4294967296 `
    -OptimalPartitionVRAM 2147483648 `
    -MinPartitionEncode 10000000 `
    -MaxPartitionEncode 50000000 `
    -MinPartitionCompute 10000000 `
    -MaxPartitionCompute 50000000

# Enable host cache types and configure MMIO space
Set-VM -VMName $VMName -GuestControlledCacheTypes $true
Set-VM -VMName $VMName -LowMemoryMappedIoSpace 134217728
Set-VM -VMName $VMName -HighMemoryMappedIoSpace 17179869184

Step 2: Inject Driver Binaries into Guest OS

GPU-P requires matching driver runtime libraries between the host OS Driver Store and the guest VM system directories^[3]:

# Locate active display driver files on Host
$HostDriverPath = (Get-ItemProperty "HKLM:\SYSTEM\CurrentControlSet\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}\0000").DriverStorePath

Write-Host "Active Host Driver Store: $HostDriverPath"

Mount the VM virtual hard disk (VHDX) or copy the files over PowerShell Direct:

1. Copy C:\Windows\System32\nv*.* (or amdkmdag*.* / igd*.*) to the guest’s C:\Windows\System32\.
2. Copy the entire driver folder from C:\Windows\System32\DriverStore\FileRepository\<DriverFolder> on the host to C:\Windows\System32\HostDriverStore\FileRepository\<DriverFolder> inside the guest VM.

Start the VM, and verify that the virtual graphics adapter initializes cleanly under Windows Display Adapters.

---

Optimizing Remote Desktop Protocol (RDP) for 60 FPS & AVC 444

Having a GPU attached to the VM is only half the configuration. By default, the Windows Remote Desktop Session Host (RDSH) uses standard CPU compression and 4:2:0 chroma subsampling, causing blurry fonts and color distortion on high-resolution displays^[4].

Configure the following Group Policy objects on the guest VM to force GPU-accelerated H.264/AVC 444 encoding:

1. Open gpedit.msc and navigate to:
   Computer Configuration > Administrative Templates > Windows Components > Remote Desktop Services > Remote Desktop Session Host > Remote Session Environment
2. Prioritize H.264/AVC 444 Graphics mode for Remote Desktop connections: Set to Enabled^[4].
3. Configure H.264/AVC hardware encoding for Remote Desktop connections: Set to Enabled (Select: Always attempt).
4. Use hardware graphics adapters for all Remote Desktop Services sessions: Set to Enabled^[4].
5. Configure image quality for RemoteFX Adaptive Graphics: Set to Enabled (Value: High).

Alternatively, apply these settings instantly via PowerShell:

$RdpRegPath = "HKLM:\SOFTWARE\Policies\Microsoft\Windows NT\Terminal Services"

If (!(Test-Path $RdpRegPath)) {
    New-Item -Path $RdpRegPath -Force | Out-Null
}

# Force Hardware GPU Acceleration
Set-ItemProperty -Path $RdpRegPath -Name "bEnumerateHWFirst" -Value 1 -Type DWord

# Enforce H.264 / AVC 444 Preferred Mode
Set-ItemProperty -Path $RdpRegPath -Name "TS_SERVER_AVC444_MODE_PREFERRED" -Value 1 -Type DWord

# Enable AVC Hardware Encoding
Set-ItemProperty -Path $RdpRegPath -Name "AVC_HARDWARE_ENCODING" -Value 1 -Type DWord

# Restart Remote Desktop Service to apply changes
Restart-Service -Name TermService -Force

---

Diagnostics and Telemetry Verification

To verify that active RDP sessions are utilizing the GPU rather than software fallback, check the Event Viewer and Task Manager telemetry:

1. RemoteDesktopServices-RdpCoreTS Event Log:
   Open Event Viewer and navigate to Applications and Services Logs > Microsoft > Windows > RemoteDesktopServices-RdpCoreTS > Operational.
   Look for Event ID 162: "The server successfully prioritized the H.264/AVC 444 graphics mode for connection."
   Look for Event ID 170: "Hardware graphics acceleration is enabled for session ID."
2. Task Manager GPU Engine:
   Under the Performance tab in Task Manager inside the session, observe the GPU 0 - Video Encode and 3D telemetry graphs during full-screen video playback or 3D viewport navigation.

---

Infrastructure Sizing for Mission-Critical RDP Farms

When designing multi-tenant remote desktop infrastructure, physical network throughput and memory bandwidth dictate real-world user density:

• Bandwidth Sizing: A single 4K 60 FPS RDP session utilizing AVC 444 hardware encoding requires between 15 Mbps and 45 Mbps of sustained, low-jitter throughput.
• Low-Latency Domestic Delivery: For organizations, software houses, and financial trading teams in Pakistan, server co-location and local peering are paramount. Deploying Dedicated Servers in Pakistan ensures single-digit round-trip latency (sub-10ms via domestic PKIX and local ISP peers), eliminating the input lag that compromises high-framerate remote interactive sessions.

---

Sources & Official References

1. Microsoft Learn — Plan for Deploying Devices using Discrete Device Assignment. Official architectural guidance for planning PCIe device assignment, host BIOS prerequisites, Access Control Services (ACS) topology, and MMIO memory boundary configuration on Windows Server Hyper-V.
2. Microsoft Learn — Deploying Graphics Devices using Discrete Device Assignment. Direct runbook detailing PowerShell cmdlets (`Dismount-VMHostAssignableDevice`, `Add-VMAssignableDevice`), host dismount procedures, and guest driver provisioning for GPUs.
3. Microsoft Learn — Add-VMGpuPartitionAdapter Cmdlet Reference (Hyper-V PowerShell Module). Official reference documentation defining GPU partitioning syntax, resource allocation boundaries (Min/Max/Optimal Partition VRAM), and paravirtualized graphics adapter configuration.
4. Microsoft Learn — Configure Graphics Hardware Acceleration for Remote Desktop Protocol (RDP). Authoritative documentation on Group Policy configurations for Remote Desktop Session Hosts, covering H.264/AVC 444 profile prioritization, hardware rasterization, and `TS_SERVER_AVC444_MODE_PREFERRED` registry parameters.

---