CPUs dedicate significant die area to smart control hardware like branch predictors, speculation, and out of order execution all necessary because general purpose software runs sequentially with unpredictable branching.GPUs restructure this tradeoff. AI workloads (matrix multiplies, dot products, etc.) are massively parallel the same operation applied to thousands of numbers simultaneously, with minimal branching in the hot path. That predictability means GPUs can afford simpler per-core control

Control - película: Ver online completa en español