DirectML is an underlying API within DirectX 12 that leverages machine learning capabilities on GPUs. Currently, DirectML supports a variety of GPUs from different manufacturers such as AMD, Intel, NVIDIA, and Qualcomm. Popular machine learning software like ONNX Runtime, Windows ML, TensorFlow, and PyTorch also utilize DirectML to access GPU processing power.
However, with the emergence of Neural Processing Units (NPUs) in personal computers, there has been a significant shift. The introduction of Intel’s Core Ultra chip has prompted Microsoft to announce DirectML support for NPUs, in addition to GPU utilization. Although there is currently no available code for real-world testing, it is expected to be open for experimentation in early 2024.
Microsoft and Intel’s plan is to enable AI features in various applications (e.g., Photoshop image editing) by leveraging NPU processing power to achieve significantly improved performance. The integration of DirectML and NPUs is anticipated to be fully realized by the year 2024.
The trend of incorporating NPU chips into personal computers is becoming more and more standardized. AMD has already started equipping certain models of their Ryzen processors with NPUs, with plans to expand further in 2023. Additionally, Qualcomm will be introducing the Snapdragon X Elite chip with NPU capabilities to target the PC market in the coming year.
TLDR: DirectML is an API within DirectX 12 that allows for machine learning tasks on GPUs. It currently supports various GPUs from different manufacturers and popular machine learning software. With the rise of NPUs in PCs, Microsoft plans to support DirectML with NPUs, alongside GPUs, by 2024. AMD is already incorporating NPUs into their Ryzen processors, and Qualcomm will release a Snapdragon X Elite chip with NPU capabilities for PCs in the near future.
Leave a Comment