Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
Windows ML provides execution providers for accelerating inference on NPU, GPU, and CPU. To learn more about accelerating inference, see Accelerate AI models.
Included execution providers
The following execution providers are included with the ONNX Runtime that ships with Windows ML:
- CPU
- DirectML (legacy)
Available execution providers
Unlike the CPU and DirectML providers, which ship in-box (see Included execution providers), the execution providers below are not included with the runtime — they're downloaded on demand.
The execution providers listed below are available on Windows 11 PCs running version 24H2 (build 26100) or greater (depending on device and driver compatibility) for dynamic download via the Windows ML ExecutionProviderCatalog APIs. To use these providers, see Install Windows ML EPs and Register Windows ML EPs. Updated versions of the execution providers are made available via Windows Update's optional nonsecurity preview releases, a.k.a. "D week releases".
The following execution providers are available to developers using Microsoft.WindowsAppSDK.ML or Microsoft.Windows.AI.MachineLearning version 2.x:
| Execution provider | Current version and release date | Upcoming version and planned release dates |
|---|---|---|
| MIGraphX (AMD) | MSIX: 1.8.57.0GPU EP: 7.2.2606.20Released: 2026 6D |
MSIX: 1.8.60.0GPU EP: 7.2.2608.30Insiders: 2026 7CGA: 2026 8D |
| NvTensorRtRtx (NVIDIA) | MSIX: 2.30.43.0Released: 2026 7D |
|
| OpenVINO (Intel) | MSIX: 1.8.80.0OpenVINO: 1.4.1Released: 2026 6D |
MSIX: 1.8.84.0OpenVINO: 1.6.1Insiders: 2026 7CGA: 2026 8D |
| QNN (Qualcomm) | MSIX: 2.2451.48.0QAIRT: 2.45.41Released: 2026 7D |
MSIX: 2.2480.49.0QAIRT: 2.48.40Insiders: 2026 7CGA: 2026 8D |
| VitisAI (AMD) | MSIX: 1.8.68.0EP: 6059 Released: 2026 7D |
MSIX: 1.8.72.0EP: 1276 Insiders: 2026 7CGA: 2026 8D |
| WebGPU (Microsoft) | MSIX: 0.2.1.0 |
Before your app uses an execution provider, please be sure to read the licenses corresponding to the execution provider.
Note
Only the current version listed for each execution provider is supported. Upcoming versions are in preview and not guaranteed to be released. Past versions are shown for release history only.
Note
Release dates are in the format of "2025 11D", or "[YEAR] [MONTH][WEEK]". "2025 11D" means it was released on the "D" week (4th week) of November 2025.
Previews are released to Windows Insiders and are often released on the "A" week (1st week) of the month. Users must be in the Windows Insider program to get those versions.
MIGraphX (AMD)
Current and upcoming releases: See the Available execution providers table above.
Execution provider info
- EpName:
"MIGraphXExecutionProvider" - Requirements:
- GPU with version 25.10.13.09 (exactly)
- This execution provider is not supported for GenAI scenarios today.
- Documentation: Documentation
- Support: Support
- License terms: Ryzen AI Licensing Information
Past releases
| Version | Windows Update release | Release notes |
|---|---|---|
| 1.8.56.0 | 2026 5D | |
| 1.8.55.0 | 2026 4D | |
| 1.8.51.0 | 2026 3D | |
| 1.8.43.0 | 2026 1D | |
| 1.8.35.0 | 2025 11D |
NvTensorRtRtx (NVIDIA)
Current and upcoming releases: See the Available execution providers table above.
Execution provider info
- EpName:
"NvTensorRtRtxExecutionProvider" - Requirements:
- NVIDIA GeForce RTX 30XX and above with minimum recommended driver version 32.0.15.5585 + Cuda version 12.5
- Documentation: Documentation
- Support: Support
- License terms: NVIDIA SOFTWARE LICENSE AGREEMENT and License Agreement for NVIDIA Software Development Kits — EULA
Past releases
| Version | Windows Update release | Release notes |
|---|---|---|
| 0.0.40.0 | 2026 6D | |
| 0.0.33.0 | 2026 5D | |
| 0.0.28.0 | 2026 4D | |
| 0.0.26.0 | 2026 3D |
OpenVINO (Intel)
Current and upcoming releases: See the Available execution providers table above.
Execution provider info
- EpName:
"OpenVINOExecutionProvider" - Requirements:
- CPU: 11th Generation Intel® Core™ processors (formerly code name Tiger Lake) or newer with at least 8GB memory
- GPU: 12th Generation Intel® Core™ processors (formerly code name Alder Lake) or newer with at least 16GB memory
- NPU: Intel® Core™ Ultra Series 1 processors (formerly code name Meteor Lake) or newer with at least 16GB memory
- Documentation: Documentation
- Support: Support
- License terms: Intel OBL Distribution Commercial Use License Agreement v2025.02.12
Past releases
For release notes of each OpenVINO version, see the OpenVINO NuGet package and OpenVINO's 2026.x and 2025.x release notes.
| Version | Windows Update release | OpenVINO version |
|---|---|---|
| 1.8.79.0 | 2026 5D | OpenVINO 1.4.0 |
| 1.8.69.0 | 2026 3D | OpenVINO 2026.0 |
| 1.8.63.0 | 2026 1D | OpenVINO 2025.4.1 |
| 1.8.26.0 | 2025 11D | OpenVINO 2025.3 |
| 1.8.18.0 | 2025 10D | OpenVINO 2025.3 |
| 1.8.15.0 | 2025 9D |
QNN (Qualcomm)
Current and upcoming releases: See the Available execution providers table above.
Execution provider info
- EpName:
"QNNExecutionProvider" - Requirements:
- Snapdragon(R) X Elite - X1Exxxxx
- Qualcomm(R) Hexagon(TM) NPU with minimum driver version 30.0.140.0 and above
- Snapdragon(R) X Plus - X1Pxxxxx
- Qualcomm(R) Hexagon(TM) NPU with minimum driver version 30.0.140.0 and above
- Snapdragon(R) X Elite - X1Exxxxx
- Documentation: Documentation
- Support: Support
- License terms: To view the QNN License, download the Qualcomm® Neural Processing SDK, extract the ZIP, and open the LICENSE.pdf file.
Past releases
For release notes of each QNN QAIRT SDK version, see Qualcomm AI Runtime (QAIRT) SDK Release Notes.
| Version | Windows Update release | QNN QAIRT SDK version |
|---|---|---|
| 2.2450.47.0 | 2026 5D | QNN 2.45 |
| 2.2420.43.0 | 2026 4D | QNN 2.42 |
VitisAI (AMD)
Execution provider info
- EpName:
"VitisAIExecutionProvider" - Requirements:
- Min: Adrenalin Edition 25.6.3 with NPU driver 32.00.0203.280
- Max: Adrenalin Edition 25.9.1 with NPU driver 32.00.0203.297
- Documentation: Documentation
- Support: Support
- License terms: Ryzen AI Licensing Information
Past releases
| Version | Windows Update release | Release notes |
|---|---|---|
| 1.8.63.0 | 2026 6D | |
| 1.8.62.0 | 2026 5D | |
| 1.8.59.0 | 2026 4D | |
| 1.8.55.0 | 2026 3D | |
| 1.8.53.0 | 2026 2D | Support for Procyon V2 Models with 13% improved score on GPT2 Machine in Turbo mode. Bug fix for disk space consumption on system C drive after restart. |
| 1.8.50.0 | 2026 1D | |
| 1.8.43.0 | 2025 11D (Windows Insiders) | |
| 1.8.31.0 | 2025 11A (Windows Insiders) | |
| 1.8.26.0 | 2025 10D | |
| 1.8.24.0 | 2025 9D |
WebGPU (experimental)
Important
WebGPU EP is experimental and requires installing experimental NuGet packages. It is not available in Windows ML 1.8.x. For full usage details, see WebGPU EP.
Current and upcoming releases: See the Available execution providers table above.
Execution provider info
- EpName:
"WebGpuExecutionProvider" - Package family name:
Microsoft.WinML.ONNX.WebGPU.EP.2 - Minimum required Microsoft.Windows.AI.MachineLearning version:
2.4.66-preview - Requirements:
- Any DirectX 12–capable GPU with up-to-date drivers (roughly 11th-gen Intel integrated graphics or NVIDIA Turing-class and newer recommended)
- Documentation: WebGPU EP
- Support: Support
- License terms: WebGPU EP for Windows App SDK License and ONNX Runtime License