AMD and Intel target a common path
AMD and Intel have announced an AMD Intel PyTorch collaboration intended to combine their x86 CPU optimization work in oneDNN. The companies say the plan will give developers a common software path for optimized AI workloads across AMD and Intel CPUs and reduce fragmentation in the AI ecosystem. oneDNN is an open-source deep learning library governed by the Unified Acceleration Foundation (UXL) that helps AI workloads run efficiently on CPUs.
Under the plan, AMD and Intel have aligned on oneDNN as the default open-source library for x86 acceleration within PyTorch. AMD says it is integrating its ZenDNN optimizations into oneDNN and has joined the UXL Foundation as a steering member. The release says the x86 Ecosystem Advisory Group recommended the move, which the companies describe as open collaboration around common technologies for x86 processors.
One location for PyTorch CPU optimizations
According to AMD and Intel, developers will have one place to access CPU optimizations for PyTorch across their processors. The companies say this should make AI applications easier to build, maintain and deploy across the x86 ecosystem. The AMD Intel PyTorch collaboration focuses on combining optimization work in oneDNN and aligning on a default open-source library for x86 acceleration within PyTorch.
Robert Hormuth, AMD corporate vice president for Architecture and Strategy, said the company is advancing an open software ecosystem intended to give developers greater choice and make high-performance AI applications easier to build. He said AMD’s work with the x86 Ecosystem Advisory Group brings the industry together around common technologies that simplify development across x86 processors.
Intel positioned the work as a way to avoid separate software paths for developers using x86 CPUs. “Developers shouldn’t have to think about separate software paths to get the best from an x86 CPU,” said Jeff McVeigh, Intel vice president and general manager for Data Center Strategic Programs. McVeigh said work with AMD through oneDNN gives the PyTorch community a consistent approach for modern processors and a route for future capabilities, including AI Compute Extensions, into the framework.
ACE is planned for future oneDNN updates
Stephen Watt, vice president in Red Hat’s Office of the Chief Technology Officer, distinguished engineer and PyTorch ambassador, described AMD and Intel standardizing x86 CPU optimizations through oneDNN in PyTorch as a step toward eliminating ecosystem fragmentation. Watt said Red Hat expects the open collaboration to make it significantly easier for developers to build, optimize and deploy AI workloads consistently across x86 architectures. He also said, “Open-source innovation thrives when the developer and operational experience is seamless and unified,” and called the collaboration a major step toward eliminating ecosystem fragmentation.
The companies are also collaborating on AI Compute Extensions (ACE), a next-generation instruction set architecture (ISA) for x86. According to the release, ACE is designed to accelerate matrix multiplication, a core mathematical operation in AI workloads. AMD and Intel say ACE will be integrated within PyTorch through future oneDNN updates. That future integration places ACE within the AMD Intel PyTorch collaboration’s stated effort to establish a common path for x86 CPU acceleration.
