AMD and Intel are combining their x86 CPU optimization work for PyTorch in oneDNN, an open-source deep learning library governed by the Unified Acceleration Foundation (UXL). The collaboration gives developers a common software path for optimized AI workloads across AMD and Intel CPUs, which the companies say reduces fragmentation in the x86 AI ecosystem.
Following a recommendation from the x86 Ecosystem Advisory Group, the two companies have aligned on oneDNN as the default open-source library for x86 acceleration within PyTorch. AMD is integrating its ZenDNN optimizations into oneDNN and has joined the UXL Foundation as a Steering Member to advance the work through open collaboration. For developers, the result is one place to access CPU optimizations for PyTorch across AMD and Intel processors, and a more consistent experience across x86 platforms. The companies say joint investment in oneDNN should make PyTorch applications easier to build, maintain and deploy on x86 systems, and speed up support for current and future x86 AI capabilities.
The companies are also collaborating on AI Compute Extensions, a next-generation instruction set architecture for x86 designed to accelerate matrix multiplication. Future support is planned for integration within PyTorch through oneDNN updates.
Key Features at a Glance
- Default library: oneDNN is the default open-source library for x86 acceleration within PyTorch, giving both vendors a shared optimization path.
- ZenDNN integration: AMD is folding its ZenDNN optimizations into oneDNN, so AMD CPU optimizations live in the same library as Intel's.
- UXL governance: oneDNN is governed by the Unified Acceleration Foundation, and AMD has joined as a Steering Member.
- AI Compute Extensions: A planned x86 instruction set for accelerating matrix multiplication, with PyTorch support to follow through oneDNN updates.
Availability
The announcement gives no release date or pricing. The AI Compute Extensions PyTorch support is described only as planned, to arrive through future oneDNN updates.
