UALink's standard facilitates communication between up to 1024 AI accelerators at speeds of 200 Gbps per lane.

The Open Compute Project Foundation (OCP) has partnered with the Ultra Accelerator Link (UALink) Consortium to enhance interconnect performance for AI and High-Performance Computing (HPC) clusters.

Earlier this month, the consortium ratified UALink 1.0, its first specification, marking a significant step in their collaborative efforts.

George Tchaparian, CEO of OCP, noted the growing demand for scalable AI infrastructure, driven by sectors such as autonomous systems and enterprise analytics.

Tchaparian emphasized that this collaboration is crucial for tackling challenges related to interconnect bandwidth and scalability necessary for advanced AI models.

The OCP community includes various stakeholders, such as hyperscale operators and technology vendors, who are committed to promoting openness and sustainability in technology development.

OCP is dedicated to designing sustainable, large-scale data center infrastructure with a focus on Open Systems for AI.

Peter Onufryk, President of the UALink Consortium, highlighted the necessity for ultra-low latency and high bandwidth to support complex data processing required by AI and HPC workloads.

Formed in October 2024, the UALink Consortium includes major tech companies like Alibaba, AMD, Apple, Google, and Microsoft, and seeks to foster an open ecosystem for data center accelerators.

Sameh Boujelbene from Dell'Oro Group stated that the alliance aims to develop advanced interconnect solutions to meet the demands of generative AI and HPC applications.

The partnership will synchronize the development efforts of both organizations to enhance interconnect specifications, addressing the bandwidth and scalability challenges posed by advanced AI models.