AMD and Intel Team Up to Make PyTorch Run Better on x86 CPUs

AMD and Intel have announced a rare partnership. They will combine their CPU optimization work for PyTorch into oneDNN, an open-source deep learning library, making it the default way to speed up PyTorch on x86 chips from both companies. PyTorch is a popular open-source framework widely used to build and train AI models. Under the plan, AMD will merge its ZenDNN optimizations into oneDNN and join the UXL Foundation. The deal was pushed by the x86 Ecosystem Advisory Group, turning competition into teamwork on the software side. The two companies are also working on AI Compute Extensions (ACE), a new instruction set that speeds up matrix math, a core part of AI training. Developers may no longer need to tune apps for each CPU separately, cutting costs and making AI apps easier to build, deploy, and move across platforms.