UALink

{{Sources exist|date=May 2025}}

{{Short description|Ultra Accelerator Link}}

Ultra Accelerator Link (UALink) is an open specification for a die-to-die interconnect and serial bus between AI accelerators. It is co-developed by Alibaba, AMD, Apple, Astera Labs,[https://www.asteralabs.com Astera Labs webpage] AWS, Cisco, Google, Hewlett Packard Enterprise, Intel, Meta, Microsoft and Synopsys.{{Cite web |title=UALink Members UCIe |url=https://www.ualinkconsortium.org/members |access-date=2024-11-01 |website=ualinkconsortium.org |language=en}}

UALink officially incorporated as an organization in 2024, and its first specification will provide interconnectivity specifically for a scalable network. The initial 1.0 version 200Gbps UALink specification, is based on the IEEE P802.3dj PHY Layer. Each system node is made up of a host and as many accelerators as needed, and is managed by one OS image. UALink Switches (ULS) connect up to 1024 accelerators within an AI 'pod', where each Accelerator is assigned a unique 10-bit routing identifier. Each UALink Switch port connects to a distinct Accelerator.{{Cite web |title=UALink 1.0 White Paper v3 |url=https://ualinkconsortium.org/wp-content/uploads/2025/04/UALink-1.0-White_Paper_v3.pdf}} The specification was due to be available to Contributor Members in 2024, and will be released to the public during the first quarter of 2025.{{Cite web |title=ABOUT UALINK |url=https://www.ualinkconsortium.org/about-ualink |access-date=2025-01-20 |website=UALink Consortium |language=en}}

It will utilize Infinity Fabric as the primary shared memory protocol. For scale out will utilize Ultra Ethernet network.