On October 20, 2026, Intel and AMD announced plans to combine their x86 CPU optimization work for PyTorch through oneDNN, an open-source deep-learning library. The goal is to give developers a common route to CPU optimizations across systems using either company’s x86 processors.
oneDNN is the shared CPU optimization route
The companies aligned on oneDNN as the default open-source library for x86 acceleration within PyTorch, following a recommendation from the x86 Ecosystem Advisory Group. That gives the collaboration a defined software focus: PyTorch optimization for x86 CPUs through one shared library.
AMD also announced that it had joined the UXL Foundation, which governs oneDNN, as a Steering Member.
AMD’s ZenDNN work and ACE plan
AMD said it was integrating its ZenDNN optimizations into oneDNN. The announcement describes that integration as work in progress.
AI Compute Extensions (ACE) is a next-generation x86 instruction set designed to accelerate matrix multiplication. AMD said PyTorch integration is planned through future oneDNN updates.
Intel’s version 2.1.0 documentation covers separate CPU and GPU paths
Intel Extension for PyTorch version 2.1.0 documents CPU optimization using AVX-512 VNNI and AMX, as well as XMX optimization for Intel discrete GPUs. The documentation says GPU features are excluded from CPU-only packages.
Those version-specific details distinguish Intel’s documented CPU and GPU paths from the announced collaboration, which focuses on x86 CPU optimization for PyTorch through oneDNN.
