A group of tech leaders from major companies such as Advanced Micro Devices, Broadcom, Cisco, Google, Hewlett Packard Enterprise (HPE), Intel, Meta, and Microsoft have joined forces to create the Ultra Accelerator Link Promoter Group. This group aims to develop a new industry standard called the Ultra Accelerator Link (UALink) to enhance communication for scale-up AI systems in data centers.
The UALink will define and establish an open industry standard that will enable AI accelerators to communicate more effectively, leading to easier integration, greater flexibility, and scalability of AI-connected data centers. This initiative is crucial for the future of AI technology, as it will provide a pathway for high-speed and low-latency communication within data centers.
Notably, Nvidia, a prominent AI chip leader, is not part of this group. However, companies like Broadcom are proud to be founding members of the UALink Consortium, emphasizing the importance of collaboration to enable large-scale AI technology implementation in data centers.
As the demand for AI compute continues to grow, having a robust, low-latency, and efficient scale-up network is essential. The UALink Promoter Group will develop a specification to define a high-speed, low-latency interconnect for scale-up communications between accelerators and switches in AI computing pods.
The UALink Consortium is expected to be official in Q3 of 2024, with the 1.0 specification available for companies that join the consortium. This initiative will advance next-generation AI/ML cluster performance and support an open ecosystem for data center accelerators, ultimately overcoming challenges in building AI supercomputers.