CONNECT WITH US

UAlink group challenges Nvidia's NVLink with open AI standard in 1Q25

Chia-Han Lee, Taipei; Levi Li, DIGITIMES Asia 0

Credit: UALink

The Ultra Accelerator Link (UALink) Consortium has officially become a legal entity, establishing itself to set a new high-speed, low-latency communication standard for AI data center servers.

Tom's Hardware reports that UALink's board members include tech giants such as AMD, Intel, Meta Platforms, Cisco, HPE, Google, Microsoft, Astera Labs, and AWS. The organization invites additional companies to join as contributing members.

According to TechRadar, these board members also established the UALink Promoter Group in May. UALink president and Intel's industry technology enablement manager, Willie Nelson, emphasized that the alliance welcomes additional members to support an open, efficient interconnect standard for AI accelerators.

UALink aims to create an industry-wide open standard for scalable AI accelerator interconnects, positioning it as a direct competitor to Nvidia's proprietary NVLink technology. NVLink facilitates communication between Nvidia GPUs and, combined with Infiniband, enables enhanced scalability.

The Infiniband standard now faces competition from the rising Ultra Ethernet technology, an open standard supported by prominent tech companies, specifically designed to challenge Nvidia's market dominance.

UALink plans to launch the UALink 1.0 specification this year, offering speeds up to 200 Gbps per channel for AI pods with up to 1,024 accelerators. The standard is initially suited for smaller clusters, connecting around eight servers, with Ultra Ethernet providing additional upgrade capacity.

The UALink standard is set for release in the first quarter of 2025, alongside the first version of Ultra Ethernet.

Industry analysts suggest that these two alliances, formed by major tech companies seeking alternatives to Nvidia's proprietary standards, aim to reduce dependence on Nvidia. With widespread industry support, these alliances could gradually establish open standards for AI data centers, challenging Nvidia's dominance.

"The work being done by the companies in UALink to create an open, high-performance, and scalable accelerator fabric is critical for the future of AI," stated Forrest Norrod, executive vice president and general manager of AMD's Data Center Solutions Group.

Beyond the Promoter Group and board members, UALink's Contributor Members include Auradine, Cadence, Lenovo, and Xconn Technologies, with Activara Technology and DreamBig Semiconductor as Adopter Members.

Samsung Electronics is expected to join UALink as a Contributor Member, following its expressed intent in June.

In the Ultra Ethernet Consortium, Steering Members include AMD, Arista, Broadcom, Cisco, Eviden, HPE, Intel, Meta, Microsoft, and Oracle. General members include major Chinese firms like Alibaba Cloud, Baidu, Huawei, ByteDance, Lenovo, and ZTE, along with global players like Qualcomm, IBM, and Dell.

Nvidia is expected to independently promote its proprietary NVLink and Infiniband technologies while working to gain further partner support. Leveraging its leadership in AI data centers, Nvidia aims to form a competitive alliance, potentially leading to a standoff between two dominant camps.