AMD and Intel Integrate PyTorch CPU Optimizations into Unified oneDNN Library

AMD and Intel are integrating their CPU optimizations for PyTorch into a unified library through the oneDNN framework. AMD has folded its ZenDNN library into the open-source oneDNN, published by the UXL Foundation and driven by Intel. The collaboration was recommended by the x86 Ecosystem Advisory Group, a joint initiative established in 2024 between the two chip manufacturers. The integration aims to simplify AI development on x86 CPUs by providing a single code path for AI inference across both AMD and Intel processors.

AMD Folds ZenDNN Library into Intel's oneDNN Framework

AMD has maintained a downstream version of oneDNN called ZenDNN, which is now moving into the open-source oneDNN. Intel originally developed oneDNN, built on top of previously deep learning libraries, before it was folded into the UXL Foundation. Intel was among the founding members of the UXL Foundation, alongside Arm, Qualcomm, Google Cloud, and others. The oneDNN backend is used for CPU inference and is built to accelerate that workload for the latest Intel and AMD chips in frameworks like PyTorch.

AMD Joins UXL Foundation as Steering Member

AMD announced it is joining the UXL Foundation as a "steering member," which "influences the future of the UXL Foundation" in exchange for a contribution of $20,000 per year. The UXL Foundation publishes the oneDNN library that now incorporates both AMD and Intel optimizations. Intel was among the founding members of the UXL Foundation, alongside Arm, Qualcomm, Google Cloud, and others.

oneDNN Delivers Performance Improvements for PyTorch Inference

When oneDNN was added to PyTorch as a backend in late 2023, it showed a 1.24x improvement with the Float32 data type, and 2.21x for the BFloat16 data type. According to the oneDNN GitHub, the library "detects hardware features at runtime" to generate optimized kernels for the platform, as well as provides low-precision data types. It also features optimized instructions for common AI and machine learning operations, such as matrix multiplication and convolution.

Unified Library Supports Cross-Platform AI Development

For developers, the move simplifies working with PyTorch across different systems, ensuring a single backend provides the proper optimizations for whatever hardware is running. Although the integration focuses on the x86 architecture, oneDNN is a cross-platform library that works on Arm and RISC-V architectures as well. The two companies have collaborated through the x86 Ecosystem Advisory Group previously, such as with the ACE extensions for x86 that were announced earlier this year.

FAQ

What did AMD and Intel announce about PyTorch CPU optimization?

AMD and Intel announced they are integrating their CPU optimizations for PyTorch into a unified library called oneDNN. AMD is folding its ZenDNN library into the open-source oneDNN framework, which is published by the UXL Foundation and driven by Intel.

How much does AMD pay to join the UXL Foundation as a steering member?

AMD contributes $20,000 per year to join the UXL Foundation as a steering member, a role that influences the future of the UXL Foundation.

What performance improvements does oneDNN provide for PyTorch?

When oneDNN was added to PyTorch as a backend in late 2023, it showed a 1.24x improvement with the Float32 data type and 2.21x improvement for the BFloat16 data type.