Windows has launched a developer preview of DirectML support for Neural Processor Unit (NPU) acceleration. This update, which is compatible with new Windows 11 devices with Intel® Core™ Ultra processors, was developed in partnership with Samsung and Intel®. It offers support for a subset of machine learning models optimized for NPU acceleration.
DirectML Unveils NPU Support in Developer Preview
Windows has rolled out a developer preview of DirectML support for Intel® AI boost. This comes with the release of DirectML 1.13.1 and the ONNX Runtime 1.17. The preview introduces Neural Processor Unit (NPU) acceleration in DirectML, the machine learning platform API for Windows.
New Features in the Developer Preview
The developer preview supports a subset of models on new Windows 11 devices with Intel® Core™ Ultra processors with Intel® AI boost. This release was developed using open source models and in close partnership with Samsung.
What is DirectML?
DirectML is a low-level API that provides a common abstraction layer for hardware vendors to expose their machine learning accelerators. It works with any DirectX 12-compatible device including GPUs and now NPUs.
Intel® Core™ Ultra Processors and Intel® AI Boost
Support for Intel® Core™ Ultra processors with Intel® AI Boost was developed in close collaboration with Intel®. The developer preview targets the NPU on new Intel® Core™ Ultra devices and runs a subset of machine learning models optimized for NPU acceleration.
Getting Started with NPU Support in DirectML
To use NPU support in DirectML, you need a Windows 11 device with an Intel® Core™ Ultra processor with Intel® AI boost, the DirectML 1.31.1 NuGet package, ONNX Runtime 1.17, and the latest NPU driver from Intel®.
Limitations of the Developer Preview
This developer preview is an early release that is intended to give you a glimpse of the future of DirectML with NPU support. It is only available for Intel® Core™ Ultra processers with Intel® AI Boost at this time and is currently only compatible with a subset of machine learning models that have been targeted for support.
From the Windows Blog
From the Windows Blog