Although AMD says its Ryzen 7040-series accelerated processing units with the company's Ryzen AI hardware engine have been shipping for well over half of a year now, there were no publicly available tools to make use of it. That changed this week, as AMD posted version 0.8 of its Ryzen AI Software Platform.
AMD's Ryzen AI for Windows is a hardware unit built into the company's Ryzen 7040-series APUs based on the Zen 4 microarchitecture. AMD's Vitis AI offers a robust toolset for AI inference, seamlessly integrating with popular frameworks like TensorFlow, PyTorch, and ONNX.
To optimize models for hardware deployment, AMD provides specialized quantization tools. The ONNX Runtime ensures efficient execution of ONNX models on AMD devices. For a streamlined development experience, AMD offers pre-configured Docker containers tailored to specific frameworks. Moreover, the platform supports integration with popular development environments such as Visual Studio and Python, making AI development easier.
AMD’s posting of Ryzen AI supporting software is timely: The company’s arch-rival Intel is set to introduce its 14th-Gen Core processors codenamed Meteor Lake this month, and one of the defining features of this CPU is indeed AI acceleration. It remains to be seen whether the blue company’s AI engine will work out of the box with programs, but at least some software developers will be able to use AMD’s Ryzen AI engine with their products in the near future.
AMD’s Ryzen XDNA AI engine is an on-die accelerator designed for lower-intensity AI inference workloads, such as audio, photo, and video processing. It aims to deliver faster response times than online services and is more power-efficient than CPU or GPU-based solutions. The engine can handle up to 4 concurrent AI streams and processes INT8 and bfloat16 instructions. AMD claims this engine is faster than Apple's M2 processor’s neural engine.