Preprint
Oct 2026
Provisioning Networks for AI Supercomputers: A Trace-Driven Study of Performance Sensitivity at Unprecedented Scale
This work presents a trace-driven study of network sensitivity for training and inference workloads at scales up to 4,096 GPUs, and combines signature-based trace decomposition with linear programming to make broad network-parameter sweeps feasible.
Tommaso Bonato, A. Rastegari, D. De Sensi et al.
· 0 citations