AMD detailed its new Quark technology in a technical blog post. Quark accelerates AI inference specifically on Ryzen AI Neural Processing Units (NPUs).
Quark enhances the ONNX-to-ONNX quantization workflow. This process allows developers to convert full-precision AI models into lower-bit versions, optimizing them for Ryzen AI hardware.
The technology enables more efficient deployment of AI models, including the YOLO family for object detection. AMD seeks significant performance gains and reduced power consumption. This efficiency is achieved compared to running the models solely on a CPU or GPU.