AMD, Intel, Google, Microsoft, Meta and others establish industry group to develop AI accelerator interconnect standard to rival NVIDIA

AMD, Intel, Google, Microsoft, Meta, Broadcom, Cisco, and Hewlett Packard Enterprise (HPE) have announced the development of an open industry standard called Ultra Accelerator Link (UALink) to enable high-speed, low-latency communications for next-generation AI accelerators in data centers.

AMD, Broadcom, Cisco, Google, Hewlett Packard Enterprise, Intel, Meta and Microsoft Form Ultra Accelerator Link (UALink) Promoter Group to Drive Data Center AI Connectivity | Business Wire
https://www.businesswire.com/news/home/20240530653602/en/AMD-Broadcom-Cisco-Google-Hewlett-Packard-Enterprise-Intel-Meta-and-Microsoft-Form-Ultra-Accelerator-Link-UALink-Promoter-Group-to-Drive-Data-Center-AI-Connectivity

By defining and establishing open industry standards that enable more effective interconnection between AI accelerators, UALink makes it easier for system OEMs, IT professionals and system integrators to achieve integration, flexibility and scalability in AI-connected data centers - providing an open, high-performance environment optimized for AI workloads.

Version 1.0 of UALink will enable the connection of up to 1024 accelerators in an AI computing pod in a reliable, scalable, low-latency network. This specification enables direct data transfer between memory attached to accelerators such as AMD's Instinct and Intel's Gaudi , improving the performance and efficiency of AI computing.

The eight companies announced this partnership will establish the UALink Consortium in the third quarter of 2024 (July to September) and release detailed specifications for UALink version 1.0.

'The work UALink companies are doing is critical to the future of AI to create an open, high-performance, scalable accelerator fabric,' said Forrest Norrod, executive vice president and general manager, Datacenter Solutions, AMD. 'Together we offer extensive experience in creating large-scale AI and high-performance computing solutions based on open standards, efficiency and robust ecosystem support. AMD is committed to contributing our expertise, technology and capabilities to this group and other open industry efforts to advance all aspects of AI technology and strengthen the open AI ecosystem.'

'UALink is an important milestone for the advancement of artificial intelligence computing. Intel is proud to co-lead this new technology and contribute our expertise to building an open and dynamic AI ecosystem,' said Sachin Katti, senior vice president and general manager of Intel's Network and Edge division. 'As a founding member of this new consortium, we look forward to a new wave of industry innovation and customer value delivered through the UALink standard. This initiative expands Intel's commitment to AI connectivity innovation, including our leadership role in the Ultra Ethernet Consortium and other standards organizations.'

Tom's Hardware, an IT news site, said, 'AMD, Broadcom, Google, Intel, Meta, and Microsoft are all developing their own AI accelerators, Cisco is manufacturing network chips for AI, and HPE is manufacturing servers. These companies are interested in standardizing their chip infrastructure as much as possible, and for that reason they are partnering in the UALink consortium. However, NVIDIA has its own infrastructure , so it is naturally not interested in co-developing UALink.' He pointed out that since NVIDIA continues to dominate in AI hardware, the launch of the UALink consortium is also a way to compete with NVIDIA.