Microsoft has announced its advancement in driving DirectML as the primary platform for developing AI on PCs if users wish to utilize their own models. After a period of development and supporting NPU since the end of last year, Microsoft is pushing for DirectML to become the standard for developing AI applications without concerning hardware differences, whether the machine is running on GPU, NPU, or solely using CPU. At this year’s Build event, it was announced that PyTorch now supports DirectML on GPU starting today and will soon support NPU as well. Additionally, AI usage on the web will begin previewing the WebNN standard backed by DirectML in the background.
DirectML itself features popular functions for AI processing such as ReLU, Sigmoid, tanh, fundamental matrix operations for running AI, or direct neural network functions like LSTM or RNN. Previously, DirectML only supported GPU, but Microsoft has now revealed plans to support both Intel’s CPU and NPU in Snapdragon X Elite soon, along with adding features to make model execution even easier.
TLDR: Microsoft is pushing DirectML as the main platform for AI development on PCs, supporting GPUs and soon NPUs, with plans to expand to CPUs and improved model execution capabilities.
Leave a Comment