In a major industry development, AMD, Amazon AWS, Astera Labs, Cisco, Google, Hewlett Packard Enterprise (HPE), Intel, Meta, and Microsoft have announced the formation of the Ultra Accelerator Link Consortium (UALink). This collaborative group aims to define a high-speed, low-latency communication standard for AI Pods and clusters, challenging existing technologies like NVIDIA's NVLink.
The UALink Consortium plans to release its first official specification, UALink 1.0, to its contributing members this year, with a general review version expected next year. The specification will enable scalable connections of up to 200 Gbps per channel for up to 1,024 accelerators within an AI Pod, enhancing communication efficiency between accelerators and switches.
The UALink Consortium's chairman, Kurtis Bowman, noted that the UALink 1.0 specification is anticipated to be published in the first quarter of 2025. This milestone is set to establish an open industry standard that will facilitate more efficient communication between AI accelerators and switches, expanding memory access to support large-scale AI models and showcasing the benefits of industry collaboration.
