Google, AMD, Meta, and Microsoft, alongside other leading technology vendors, have introduced a new industry standard for AI connectivity in data centers. The Ultra Accelerator Link (UALink) aims to enhance performance and deployment flexibility in AI computing clusters.
UALink is designed for accelerators found on GPUs, facilitating more efficient interconnections among hardware that powers AI training and inference workloads. The initial version, UALink 1.0, will allow data center operators to connect up to 1,024 accelerators within a single computing pod, with formal adoption slated for later this year.
Key players such as AMD, Broadcom, Cisco, Intel, and HPE have joined in establishing this open industry standard. The UALink standard will enable data centers to scale their computing resources dynamically, adding capacity without interrupting ongoing workloads.
“Ultra-high performance interconnects are becoming increasingly important as AI workloads continue to grow in size and scope,” remarked Martin Lund, Executive Vice President of Cisco’s Common Hardware Group. “Together, we are committed to developing UALink, which will be a scalable and open solution to help overcome some of the challenges with building AI supercomputers.”
Forrest Norrod, General Manager for AMD’s Data Center Solutions Business Group, emphasized the importance of this collaborative effort. “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.
These companies are part of the Ultra Ethernet Consortium (UEC), an industry group dedicated to Ethernet-based networking cooperation. J Metz, UEC’s Chair, highlighted the industry's rapid adaptation to AI and high-performance computing challenges. “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,” Metz noted.
Despite broad support, Nvidia is notably absent from the list of companies endorsing UALink, as it continues to use its proprietary NVLink technology for GPU interconnections.
This new standard is poised to significantly influence the future of AI and data center operations, promoting an open, scalable, and efficient framework for advancing AI technologies.
