facebook rss twitter

AMD Announces GPU-Accelerated Machine Learning Training on Windows 10


Quick Link: HEXUS.net/qaemhs

Add to My Vault: x

AMD has announced support for GPU-accelerated machine learning training workflows on Windows 10, which allows users with AMD hardware – from software engineers to students – to access ML training workflows and hone their skills on the same PCs they use for day-to-day tasks.

The update follows Build 2020, where Microsoft announced it is adding GPU-accelerated compute support to the Windows Subsystem for Linux (WSL), allowing the use of command-line Linux applications side-by-side with native Windows 10 applications. 

Up until now, the Microsoft DirectML API only enabled GPU-accelerated ML inference on any DirectX 12 compatible GPU, including AMD Radeon and Radeon Pro graphics cards. Now, machine learning training workflows can also be GPU-accelerated on Windows 10, and Microsoft is also working to integrate DirectML into the most used machine learning tools, libraries and frameworks.

Microsoft has launched a preview of this feature for members of the Windows Insider Program, offering a package that includes a TensorFlow build with a DirectML backend, along with TensorFlow tutorial models.

To support the Microsoft preview, AMD released a preview Radeon Software Adrenalin 2020 Edition driver that enables support for DirectX 12 compatible GPU-acceleration within the WSL.

The driver is compatible with AMD Radeon graphics, Radeon Pro graphics, Ryzen and Ryzen PRO Desktop Processors with Radeon Vega graphics, as well as Ryzen and Ryzen PRO Mobile Processors with Radeon Vega graphics. For a complete list of compatible AMD products, please refer to the release notes.

To check out GPU-accelerated ML training in action using an AMD Ryzen Microsoft Surface Edition processor in the Microsoft Surface Laptop 3, click here. And, to experience the preview on your AMD-powered system, download AMD’s preview driver and follow the directions outlined in Microsoft’s getting started documentation.