According to reports, Intel, Google, Microsoft, Meta, and other tech giants have announced the establishment of a new industry organization called "Ultra Accelerator Link (UALink) Promotion Group".

The group aims to establish industry standards, lead the development of connectivity components between AI accelerator chips in data centers, and challenge Nvidia's dominance in AI accelerators. In addition to the aforementioned companies, members of the organization also include AMD, HP Enterprise (HPE), Broadcom, Cisco, etc. Nvidia and Arm have not yet participated.

The UALink promotion team is proposing a new industry standard to connect AI accelerator chips in an increasing number of servers. In a broad sense, AI accelerators refer to GPUs and other custom designed solutions used to accelerate the training, fine-tuning, and operation of AI models.

According to reports, the first standard version proposed by UALink, UALink 1.0, will connect up to 1024 GPU AI accelerators to form a computing "pod", which refers to one or more racks in a server. According to the UALink promotion team, based on "open standards" including AMD's Infinity Fabric, UALink 1.0 will allow direct loading and storage between the memory attached to AI accelerators, and overall improve speed while reducing data transfer latency compared to existing interconnect specifications.

The group stated that it will create an alliance, the UALink Alliance, in the third quarter to oversee the future development of UALink standards. UALink 1.0 will be available to companies joining the alliance during the same period, and the updated specification UALink 1.1 with higher bandwidth is planned to be launched in the fourth quarter of 2024.

Currently, Nvidia has not commented on this matter, but it is clear that the company is not keen on supporting UALink. Analysis suggests that Nvidia has already provided its own proprietary interconnect technology for GPU links within data center servers, so it may not be willing to support specifications based on competitor technologies.

The copyright of this article belongs to the original author. The reprint of the article is only for the purpose of disseminating more information. If the author's information is marked incorrectly, please contact us immediately to modify or delete it. Thank you for your attention!