Intel and AMD to unify x86 PyTorch optimization in oneDNN
The two chipmakers will make oneDNN the default open-source library for x86 acceleration in PyTorch, under the UXL Foundation's governance.
Intel and AMD announced on October 21, 2026 that they will jointly consolidate x86 CPU optimization technology for the PyTorch AI framework around the open-source oneDNN library, according to techm.kr. Under the plan, oneDNN becomes the default open-source library for x86 acceleration within PyTorch.
The arrangement means developers will be able to reach PyTorch CPU optimization technology for both companies' processors in one place, rather than through separate vendor-specific paths.
Governance of the shared library sits with the UXL Foundation. Intel and AMD said the consolidation follows a recommendation from the x86 Ecosystem Advisory Group.
The two companies are also collaborating on AI Compute Extensions, a next-generation x86 instruction set architecture designed to accelerate matrix multiplication. They plan to support it in PyTorch through future oneDNN updates.
Background
Before the October 2026 announcement, AMD was already integrating its ZenDNN optimization technology into oneDNN and had joined the UXL Foundation as an operating member to expand the collaboration, according to techm.kr.
Quick answers
What is oneDNN used for in PyTorch?
Under the new arrangement, oneDNN is the default open-source library for x86 CPU acceleration within PyTorch, covering both Intel and AMD processors.
Who governs oneDNN?
The library is governed by the UXL Foundation, of which AMD is an operating member.
What are AI Compute Extensions?
A next-generation x86 instruction set architecture designed to accelerate matrix multiplication, which Intel and AMD plan to support in PyTorch through future oneDNN updates.