AMD and Intel announced a collaboration on October 20 to optimize PyTorch for x86 CPUs. The companies will integrate their respective optimization efforts into the open-source library oneDNN to provide a unified software path for developers across both CPU architectures.
As part of this initiative, AMD will integrate its ZenDNN optimizations into oneDNN. Furthermore, AMD has joined the UXL Foundation as a member of the steering committee. This collaboration is facilitated by the x86 Ecosystem Advisory Group to foster cooperation in the CPU software landscape.
The companies are also developing AI Compute Extensions (ACE), a next-generation instruction set architecture for x86. ACE focuses on accelerating matrix multiplication, which is critical for AI model training and inference. Future updates will integrate these capabilities into PyTorch via oneDNN.
This unified approach aims to reduce fragmentation in the x86 AI ecosystem. Developers, cloud providers, and enterprise users are expected to benefit from more consistent PyTorch performance across x86 servers, workstations, and AI PCs, while lowering development and maintenance costs.
