Researchers Model Brain Simulation Communication Efficiency

Bruno Golosio et al. (2025) model communication strategies for brain-scale neural simulations on an NVIDIA A100‑equipped Leonardo Booster supercomputer, analyzing point-to-point and collective messaging. They estimate Leonardo's 3,456 nodes could simulate about 3.5×10^13 connections—roughly 10% of the human cortex under random connectivity—and recommend a hybrid communication design, implying fruitfly-scale simulations are near-term feasible while whole-human simulations remain out of reach.