Installation

Installation#

This page consolidates all requirements and installation steps for running Windows ML on Ryzen AI PCs.

Prerequisites#

Requirement

Version or Notes

Windows

Windows 11 24H2 (build 26100) or greater

Ryzen AI NPU

Supported processor with NPU. See supported configurations in the release notes.

Visual Studio (for C++)

Visual Studio 2022, latest version. Ensure Desktop Development with C++ is installed.

Visual Studio Code (optional)

For model conversion using AI Toolkit extension

Python (for Python examples)

3.10 to 3.12

C++ (for C++ examples)

C++20 or later

For the complete list of supported Windows versions, refer to Windows App SDK support.

Installation#

Follow these steps in order:

  1. Install NPU drivers: Follow the RAI installation instructions. Download and install the NPU driver (version 32.0.203.280 or newer) from the AMD Ryzen AI driver page.

  2. Install Windows App SDK: Windows ML is included as part of the Windows App SDK. Install the version required by your branch/sample from Windows App SDK downloads.

  3. Verify installation: For Python, ensure the wasdk-microsoft-windows-ai-machinelearning package is installed (e.g., via pip install) and matches the Windows App SDK version used by your sample branch. Run conda list | findstr wasdk to verify.

Note

The Windows ML VitisAI Execution Provider (EP) NuGet packages are available for download:

Version

Package Download

Release 2026.5D

AMD.NPU.WinML.VitisAI.EP_2026_5D.nupkg

Release 2026.7D

AMD.NPU.WinML.VitisAI.EP_2026_7D.nupkg

Key Features of Windows ML#

  • Dynamically downloads latest EPs: Compatible execution providers are downloaded from the Microsoft Store on demand

  • Shared Windows-wide ONNX Runtime: Reduces application size; no need to bundle ORT

  • Smaller downloads and installs: EPs are shared across applications

  • Broad hardware support: Works across CPUs, GPUs, and NPUs from different vendors via ONNX Runtime

Windows ML setup verification#

Install the required Python packages in the conda environment winml_env

conda create -n winml_env python==3.11
conda activate winml_env
git clone https://github.com/amd/RyzenAI-SW.git
cd <RyzenAI-SW>\WinML\CNN\ResNet
pip install --pre -r .\requirements.txt

Check the installed wasdk Python version and install same version of Windows App SDK:

conda list | findstr wasdk

Download the Windows App SDK that matches the wasdk version to ensure the Windows ML execution providers work as expected.

curl -L -o windowsappruntimeinstall-x64.exe "https://aka.ms/windowsappsdk/2.3/2.3.1/windowsappruntimeinstall-x64.exe"
windowsappruntimeinstall-x64.exe --quiet

Getting Started Examples#

The following examples provide step-by-step instructions to help you get started with Windows ML on AMD Ryzen AI PCs. These examples cover CNN, Transformer, and LLM model deployment using both C++ and Python APIs.

For more details about model deployment using Windows ML, see the Model Deployment using Windows ML documentation.

Contents