IT Home reported on December 18 that Microsoft recently announced that it will add NPU support to the DirectML machine learning Xi framework early next year, so as to adapt to chips with built-in NPU such as Intel Core Ultra.
▲ Source: Microsoft's official press release (the same below)
In 2021, Microsoft launched the DirectML framework belonging to the DirectX family, which focuses on "machine learning Xi", which can directly access the GPU for deep learning Xi, and can help a series of scenarios where AI is involved in computing, such as "image denoising", "game pre-rendering", and "ray tracing".
Microsoft said that with DirectML support for NPUs, developers can use the on-chip NPU to run AI models through APIs such as the cross-platform inference engine "ONNX Runtime", thereby improving model performance.
IT Home noted that Microsoft also said that DirectML will "directly support" the Core Ultra chip early next year, and the NPU built into the relevant chip will become "the first NPU on the Windows platform to support DirectML".
▲ Intel's previously released Core Ultra series processors, source Intel's official website
In addition to Intel Core Ultra, new generation chips from Qualcomm, AMD and other manufacturers also have built-in NPUs, allowing AI computing at relatively low power consumption, thereby freeing up CPU and GPU computing resources.