Microsoft, Meta, Google Launch New AI Connectivity Standard for Data Centers

Google, AMD, Meta and Microsoft, along with other technology vendors, have launched a new industry standard for AI connectivity in data centers.

The Ultra Accelerator Link (UALink) is designed to improve performance and deployment flexibility in AI computing clusters housed in data centers.

UALink applies to accelerators found on GPUs, enabling hardware powering AI training and inference workloads to interconnect with one another more efficiently.

Version 1.0 of the standard will enable data center operators to connect up to 1,024 accelerators in a single computing pod. It is set to be formally adopted later this year.

AMD, Broadcom, Cisco, Intel and HPE also signed on to form the open industry standard.

The companies said the UALink standard will enable data centers to add computing resources to a single instance, allowing them to scale capacity on demand without disrupting ongoing workloads.

“Ultra-high performance interconnects are becoming increasingly important as AI workloads continue to grow in size and scope,” said Martin Lund, executive vice president of Cisco’s common hardware group. “Together, we are committed to developing the UALink which will be a scalable and open solution available to help overcome some of the challenges with building AI supercomputers.”

Related:AI Workloads to Double Data Center Power Demand by 2026

Forrest Norrod, AMD’s general manager for AMD’s data center solutions business group, said the work being done by companies in the UALink to create an open, high performance and scalable accelerator fabric is critical for the future of AI.

“Together, we bring extensive experience in creating large-scale AI and high-performance computing solutions that are based on open standards, efficiency and robust ecosystem support,” Norrod said.

The companies that adopted the UALink standard are members of the Ultra Ethernet Consortium (UEC), an industry group supporting cooperation around Ethernet-based networking.

“In a very short period of time, the technology industry has embraced challenges that AI and high-performance computing have uncovered,” said J Metz, UEC’s chair. “Interconnecting accelerators like GPUs requires a holistic perspective when seeking to improve efficiencies and performance. At UEC, we believe that UALink’s scale-up approach to solving pod cluster issues complements our own scale-out protocol and we are looking forward to collaborating together on creating an open, ecosystem-friendly, industry-wide solution that addresses both kinds of needs in the future.”

A notable absence among the companies who pledged themselves to the standard is Nvidia, which uses its own NVLink to interconnect GPUs.

Related:Meta Building New $800M AI Data Center in Alabama, Second in State