AMD Joins UXL as Intel and AMD Standardize oneDNN for PyTorch

Overview

• AMD became a steering member of the UXL Foundation on October 20, 2026, and together with Intel committed to making oneDNN the default open-source library for x86 acceleration in PyTorch.
• As part of the agreement, AMD is folding its ZenDNN CPU optimizations into oneDNN so those performance improvements will live in a shared library rather than in a vendor-specific stack.
• oneDNN already supports major machine-learning runtimes such as PyTorch, TensorFlow, ONNX Runtime and OpenVINO, so making it the PyTorch x86 default can affect performance and workflows across the broader AI tooling stack.
• AMD will continue to develop its ROCm GPU software and its own PyTorch workstreams while oneDNN’s support for AMD GPUs remains experimental via SYCL, HIP and ROCm.
• The change follows multi-year cooperation in the x86 Ecosystem Advisory Group and places oneDNN under the UXL Foundation, which operates inside the Joint Development Foundation of the Linux Foundation to manage cross-vendor collaboration.