Data movement bottlenecks to large-scale model training: Scaling past 1e28 FLOP | Hacker News Reader