Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about SONiC's approach to building scale-up AI clusters in this 35-minute conference talk from the Open Compute Project. Explore the increasing demand for scale-up clusters that tightly couple large numbers of GPUs to minimize communication overhead for training and inference workloads. Discover the three major categories of approaches for building scale-up GPU clusters: proprietary interconnect protocols, standard Ethernet, and UAL fabrics. Understand how the SONiC community focuses on enabling and optimizing Ethernet-based scale-up AI clusters, which provide openness, flexibility, reliability, and ecosystem compatibility. Gain insights from leading hyperscalers regarding their technical requirements and examine key areas of ongoing community discussion. Learn about SONiC's collaboration with the Ultra Ethernet Consortium (UEC) to address emerging needs in AI cluster infrastructure and discover opportunities to join the community effort in building these advanced cluster systems.
