AI Discussions

[Yilong]

I ran Microsoft's official DirectML sample and found that DirectML can only detect the 890M graphics card but not the NPU. Are there any plans to support this? I would really like to have related support so that I can conveniently develop applications on the NPU.

[Uday_Das]

NPU is supported through the Vitis AI Execution provider as of today (not through DirectML) https://ryzenai.docs.amd.com/en/latest/inst.html

[Yilong]

Is there any plan to support directml? Both Intel and Qualcomm npu can be used on Windows through directml

[Benstock55]

Yeah, that would be awesome! DirectML support could really unlock more power from Intel and Qualcomm NPUs on Windows.

[Benstock55]

Do you have any plans to add support for DirectML? Since both Intel and Qualcomm NPUs can be leveraged on Windows through DirectML, it could enhance performance and efficiency for compatible hardware.

[yunhao2]

Hello,

As far as I know directml can only expend dx devices and can only detect my discrete and integrated graphics cards, how can I make inference run on Intel NPUs directly through directml ?
Thanks a lot.

[jack5253]

The Ryzen AI 9 HX 370 NPU does not currently have native support for DirectML, which primarily supports GPU-accelerated machine learning workloads. DirectML (Direct Machine Learning) is optimized for GPUs and certain accelerators like Intel’s integrated AI or discrete GPUs from NVIDIA and AMD. As of now, DirectML recognizes the 890M graphics card, but it doesn't directly interface with NPUs like the one on Ryzen AI chips.

There isn't clear public information yet on specific plans to integrate NPU support into DirectML for this model. AMD and Microsoft may need to collaborate to create the necessary drivers or interfaces to expose the NPU to DirectML in the same way that GPUs are supported. In the meantime, you might want to explore using custom or proprietary APIs for Ryzen’s NPU, or alternative libraries that can interface directly with the NPU hardware for machine learning tasks. Keep an eye on official documentation from AMD and Microsoft for any future announcements regarding DirectML support for NPUs.

[jack5253]

Specific intentions to incorporate NPU functionality into DirectML for this architecture are not yet well known to the general public. In order to expose the NPU to DirectML in the same manner as GPUs are supported, AMD and Microsoft might need to work together to provide the required drivers or interfaces. For machine learning activities, you may want to look into using alternative libraries that can interact directly with the NPU hardware or unique or private APIs for Ryzen's NPU. For any upcoming news on DirectML support for NPUs, keep a watch on the official documentation from AMD and Microsoft.

[nysolutions]

AMD has not announced plans to integrate DirectML support for the NPU, future software updates or collaborations with Microsoft could potentially enable this functionality.