AMD and Intel announced a collaboration to unify x86 CPU optimizations in PyTorch through the oneDNN library. The AMD Intel oneDNN PyTorch initiative aims to give developers a single, optimized software path for AI workloads on processors from both companies.
oneDNN is an open-source deep learning library governed by the Unified Acceleration Foundation (UXL). AMD will integrate optimizations from its ZenDNN fork directly into oneDNN and has become a steering member of the UXL Foundation. Both companies have set oneDNN as the default open-source library for x86 acceleration in PyTorch.
Why the AMD Intel oneDNN collaboration matters
Until now, developers running AI inference on CPUs had to deal with separate paths: oneDNN for Intel and ZenDNN for AMD. The unification reduces this fragmentation, making it easier to build, maintain and deploy AI applications on x86 hardware.
The recommendation came from the x86 Ecosystem Advisory Group. The companies are also collaborating on AI Compute Extensions (ACE), a new ISA for x86 focused on matrix multiplication — a core operation in AI workloads — with support planned via future oneDNN updates in PyTorch.
Robert Hormuth, corporate vice president of Architecture and Strategy at AMD, said the company is advancing an open software ecosystem that gives developers more choice. Jeff McVeigh of Intel highlighted that developers should not have to think about separate software paths to get the best from an x86 CPU.
This is a rare collaboration between historical rivals in the processor market and signals prioritization of interoperability in the CPU AI ecosystem, especially relevant for servers, workstations and scenarios where GPUs are not the only inference option.
Sources: Official AMD Newsroom announcement (October 20, 2026); coverage by Tom's Hardware and Phoronix.
Transparency: This content was created, edited or reviewed with the assistance of artificial intelligence. The information was cross-checked with public posts on X and sources available on the internet. Consult the original sources to verify the full context.
By GeekikiBot
