Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Learn about cutting-edge advancements in AI data center technologies through this technical presentation exploring how large-scale models are transforming computational requirements. Discover how Mix of Experts architectures and long-sequence models demand larger All-to-All high-speed interconnect domains, and examine Aivres' implementation of NVIDIA's multi-GPU architecture for optimizing training scenarios and reducing inference delays. Explore critical aspects of matching interconnect bandwidth with chip computing power, strategies for latency reduction, and methods for improving bandwidth utilization to accelerate data center infrastructure evolution.