UALink

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,[1] AWS, Cisco, Google, Hewlett Packard Enterprise, Intel, Meta, Microsoft and Synopsys.[2]

The UALink consortium officially incorporated as an organization and electronics industry consortium in 2024, for promoting and advancing UALink. 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.[3] 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.[4]

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

References

  1. ^ Astera Labs webpage
  2. ^ "UALink Members UCIe". ualinkconsortium.org. Retrieved 2024-11-01.
  3. ^ "UALink 1.0 White Paper v3" (PDF).
  4. ^ "ABOUT UALINK". UALink Consortium. Retrieved 2025-01-20.

See also

This article is issued from Wikipedia. The text is available under Creative Commons Attribution-Share Alike 4.0 unless otherwise noted. Additional terms may apply for the media files.