Intel and AMD Simplify x86 CPU AI Development

According to announcements made on October 21, Intel and AMD are collaborating to consolidate their respective x86 CPU optimizations for PyTorch into oneDNN, an open-source deep learning performance library managed by the Unified Acceleration Foundation (UXL). The initiative aims to improve AI workload efficiency on CPUs, reduce software fragmentation across the x86 ecosystem, and simplify the development, maintenance, and deployment of PyTorch applications across different x86 platforms.

Following recommendations from the x86 Ecosystem Advisory Group, Intel and AMD have agreed to designate oneDNN as the default open-source library for x86 acceleration in PyTorch. AMD is integrating its ZenDNN optimizations into oneDNN and has joined the UXL Foundation as a Steering Committee member to support further development through open collaboration.

For developers, the collaboration provides a unified software path for accessing PyTorch CPU optimizations across Intel and AMD processors. By consolidating optimization efforts within oneDNN, the two companies aim to deliver a more consistent development experience across x86 systems while reducing the complexity of supporting different processor architectures. The initiative is also expected to strengthen software support for current and future x86 AI capabilities.

Intel and AMD are also working together on AI Compute Extensions, a next-generation instruction set architecture designed for x86 processors to accelerate matrix multiplication. These capabilities are expected to be integrated into PyTorch through future oneDNN updates, potentially enabling further improvements in CPU-based AI performance.

Jeff McVeigh, Intel Vice President and General Manager of Strategic Projects for Data Center, said the collaboration aims to eliminate the need for developers to follow separate software paths to maximize x86 CPU performance. He added that a unified approach would benefit the PyTorch community and make it easier to integrate future performance enhancements, including AI Compute Extensions, into the framework.

Robert Hormuth, AMD Corporate Vice President of Architecture and Strategy, emphasized the company's commitment to open software ecosystems and greater developer choice. He noted that collaboration with the x86 Ecosystem Advisory Group would help promote broader industry cooperation around common technologies and simplify development across x86 processors.

Stephen Watt, Vice President and Distinguished Engineer in the Office of the CTO at Red Hat and a PyTorch Ambassador, welcomed the initiative, highlighting the importance of consistent development and operational experiences in open-source innovation. He said Intel and AMD's joint efforts to standardize x86 CPU optimizations through oneDNN in PyTorch represent an important step toward reducing ecosystem fragmentation. The collaboration is expected to help developers build, optimize, and deploy AI workloads more easily and consistently across different x86 architectures.