AMD and Intel partner to unify AI optimizations for x86 CPUs

AMD and Intel have announced an agreement to consolidate their development efforts for x86 CPU artificial intelligence optimizations. The two semiconductor manufacturers will merge their adjustments for the PyTorch framework into the open-source deep-learning library oneDNN.

Under this initiative, AMD will integrate its existing ZenDNN optimizations into oneDNN, establishing it as the central standard path for x86 acceleration under PyTorch. This move follows a recommendation from the x86 Ecosystem Advisory Group, an entity established by both companies in 2024. As part of the cooperation, AMD is joining the Unified Acceleration Foundation (UXL) as a leading member.

The primary objective is to create a unified software path for PyTorch computations. This allows models developed on one manufacturer's cores to run on a competitor's chips without requiring manual code rewrites or the maintenance of separate, manufacturer-specific software paths. While the collaboration aims to simplify deployment and maintenance for developers, the companies noted that hardware-specific instructions and workloads will still influence final performance results.