AMD and Intel Align PyTorch CPU Optimization Around oneDNN
AMD and Intel will converge x86 CPU optimizations for PyTorch in oneDNN, with AMD integrating ZenDNN work.
AMD and Intel announced October 20 that they are combining their x86 CPU optimization work for PyTorch in oneDNN, the open-source deep-learning library governed by the Unified Acceleration Foundation.
The companies aligned on oneDNN as the default open-source acceleration path for x86 within PyTorch. AMD plans to integrate ZenDNN optimizations into oneDNN and joined the UXL Foundation as a steering member. The effort is intended to reduce separate software paths for AMD and Intel processors while preserving hardware choice.
They also said they are collaborating on AI Compute Extensions for future x86 matrix-multiplication acceleration, with planned PyTorch support through oneDNN updates. The announcement establishes a software roadmap; it does not provide deployment dates for those future instructions or independently benchmark production workloads.
Official source · Source reviewed October 21, 2026.