Artificial intelligence (AI) is transforming the way we interact with technology, enabling new and enhanced experiences in a variety of scenarios. DirectML, a development platform, offers programmers the ability to bring their AI innovations to Windows and use local hardware acceleration to scale across a wide range of Windows devices. Today, we are pleased to share how DirectML continues to expand platform support for Neural Processing Units (NPUs) on Copilot+ and WebNN PCs.
We are excited to announce that DirectML is now compatible with Copilot+ PCs, which run on Qualcomm’s Snapdragon® X Elite computing platform. These Copilot+ PCs offer exceptional performance and energy efficiency, enabling amazing AI experiences on Windows. DirectML, a key component of Windows Copilot Runtime, now allows developers to target these machines to scale AI on Windows.
Upendra Kulkarni, Vice President of Product Management for Computing Software at Qualcomm, shares Microsoft’s excitement: “With Snapdragon X Elite, we introduced an industry-leading NPU with 45 TOPS of AI performance and incredible energy efficiency. DirectML is a developer-friendly ML programming interface that leverages a familiar DirectX API framework. By supporting DirectML on our NPU, developers can now easily access its phenomenal capability and transfer their models from GPU to NPU with minimal effort. We have collaborated extensively with Microsoft to optimize DirectML for NPU and maximize hardware performance. We are thrilled to co-announce this developer preview program.”
To get started with DirectML on Copilot+ PCs, follow these steps. First, make sure you have the correct versions of DirectML, ONNX Runtime (ORT), Windows, and the minimum Qualcomm® Hexagon NPU driver. Then, create a developer account with Qualcomm and download the latest driver. With the driver package installed and minimum versions of DirectML, ORT, and Windows enabled, you can test a Super-Resolution ESRGAN model to build a C++ application that works on the new Copilot+ PCs.
DirectML is rapidly expanding and we are simplifying how developers can utilize various local accelerators, hardware variations, and frameworks on Windows. However, there are some current limitations. C++ developers must rely on the DirectML 1.15.2 redistributable package within their applications and NPU selection through the Python API for ONNX Runtime is not yet available. It is ensured that certain models work on the Qualcomm® Hexagon NPU and developers are recommended to use the exact versions of ONNX mentioned.
Additionally, DirectML unlocks NPUs for web-based machine learning with the new WebNN NPU Developer Preview launched at Build, which also allows web developers to leverage the power of NPUs on Windows devices. WebNN, an emerging web standard, defines how to run machine learning models in the browser and offers new possibilities to bring AI innovations to the web. It integrates hardware acceleration APIs like DirectML, enabling websites to leverage the GPU or NPU on the user’s PC.
This expansion of DirectML through NPUs with WebNN has been made possible thanks to close collaboration with hardware partners like Intel, which provides support for Intel® Core™ Ultra processors with Intel® AI Boost. This integration is expected to give web developers the tools necessary to create and run innovative machine learning features in web applications, benefiting users with faster, smarter, and more engaging web applications.
To get started with the WebNN developer preview, you will need the latest NPU drivers and a compatible Windows 11 version, along with specific versions of Microsoft Edge browser enabled for WebNN. Developers can run ONNX models in the browser with hardware acceleration and minimal changes to the code. Detailed instructions on usage and links to specific models are available on the WebNN Developer Preview website.
DirectML remains committed to expanding support on Copilot+ PCs and frameworks like WebNN and ONNX Runtime. We invite developers to provide feedback and report issues through the DirectML GitHub repository. This is just the beginning of our mission to enable AI on the web. Stay tuned for future updates and new capabilities in this exciting area.
Referrer: MiMub in Spanish
