Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn to design and analyze large-scale GPU superclusters capable of supporting up to 1 million GPUs through this 20-minute conference talk from the Open Compute Project. Discover the Super Cluster Builder tool and methodology that addresses the critical need for evaluating compute infrastructures as AI workloads scale rapidly. Explore both fully-connected and hierarchical design approaches while gaining insights into architectural, performance, and cost trade-offs. Master the modeling of interconnect configurations using various physical layers including copper and photonic technologies, and learn to visualize topology diagrams effectively. Understand how to quantify infrastructure components such as switches and cables, and conduct detailed complexity analysis covering hop-counts, peer-to-peer bandwidth, cross-sectional bandwidth, and latency metrics essential for accurate performance prediction. Gain expertise in total cost of ownership (TCO) analysis by correlating infrastructure complexity with cost drivers, including hardware density and cable specifications, presented by experts from Samsung and Cal Poly Pomona.
