Skip to content

UALink

An open specification for accelerator-to-accelerator links

Visit official site

What UALink is

The standard is described as an open industry consortium effort to define communication between AI accelerators such as GPUs.

What you can do with it

  • Enable direct load, store and atomic operations
  • Link hundreds of accelerators with low latency and high bandwidth
  • Apply simple load and store semantics with software coherence

Application scenarios

  • A cloud provider connects multiple GPUs for compute services
  • A system OEM integrates the interface into servers
  • An accelerator developer builds a new generation of AI chips

Pros & cons

✓ What we like

  • Simple load and store semantics with software coherence
  • Members cover cloud, OEM, accelerator, switch and IP fields

! What to watch out for

  • Developed as an open industry standard specification
  • Extra data-centre AI connectivity models are still in development

FAQ

What does the group develop?

Technical specifications for accelerator-to-accelerator communication.

What does version 1.0 draw on?

The experience of founding members in accelerators and switches.

Last reviewed: 2026-11-15

More AI coding tools tools

View all →

How we review