AMD and Intel merge PyTorch tools into oneDNN library
AMD will fold its ZenDNN optimization technology into Intel's oneDNN, which becomes PyTorch's default x86 acceleration library.
AMD and Intel announced on October 23, 2026 that they will merge their separately developed PyTorch optimization work into the open-source oneDNN library, according to digitaltoday.co.kr. Under the plan, AMD will integrate its ZenDNN optimization technology into Intel's oneDNN.
With the merger, oneDNN becomes the default open-source library for x86 acceleration in PyTorch. The companies said the goal is to make AI development easier on x86 CPUs.
The two chipmakers recently co-developed the AI Compute Extensions (ACE), a next-generation x86 instruction set extension for fast matrix multiplication. ACE support will be included in future oneDNN updates, the companies said.
Nvidia released Blackwell GPUs, Grace CPUs and RTX Spark laptops with up to 128GB of unified memory, and is expanding its Vera CPU aimed at the agent market. Apple's Mac mini and Mac Studio have meanwhile emerged as small computers for running local AI models and agents.
Quick answers
What is oneDNN used for?
oneDNN becomes the default open-source library for x86 acceleration in PyTorch, following the merger of AMD's and Intel's optimization work.
What happens to AMD's ZenDNN?
AMD will integrate its ZenDNN optimization technology into Intel's oneDNN.
Will oneDNN support ACE?
Yes. AMD and Intel said ACE support will be included in future oneDNN updates.