AMD and Intel collaborate on unified AI development path for x86 processors

AMD and Intel have announced a rare collaboration to unify AI development across x86 processors by integrating their respective optimizations into the open-source oneDNN library. This partnership aims to provide developers with a consistent software path for PyTorch on both AMD and Intel CPUs. AMD will integrate its ZenDNN optimizations into oneDNN and join the Unified Acceleration Foundation as a member of its steering committee. The collaboration, facilitated by the x86 Ecosystem Advisory Group, seeks to reduce fragmentation in the x86 AI ecosystem and lower development and maintenance costs for developers, cloud providers, and enterprise users. Additionally, both companies are advancing the AI Compute Extensions (ACE), a new instruction set architecture for x86 that will enhance matrix multiplication performance for AI training and inference. These updates are expected to be integrated into PyTorch through oneDNN in the future.