AMD and Intel unify PyTorch CPU optimizations in oneDNN

AMD and Intel are consolidating their x86 CPU optimization work for PyTorch into oneDNN, an open-source deep learning library governed by the UXL Foundation. AMD is folding its ZenDNN optimizations into oneDNN and joining UXL as a Steering Member, while both companies work on AI Compute Extensions for matrix multiplication.

• oneDNN becomes the default PyTorch CPU acceleration library for AMD and Intel
• AMD integrates ZenDNN into oneDNN and joins the UXL Foundation
• Joint work on AI Compute Extensions ISA for matrix operations
• Goal is less x86 AI fragmentation and easier app portability