fly-brain-research (framework benchmark)

Runs the same FlyWire v783 whole-brain LIF model on six simulator backends - Brian2 CPU, Brian2CUDA, PyTorch, NEST GPU, GeNN and Brian2GeNN - and times them on a shared grid of durations and trial counts, keeping setup cost separate from simulation cost so compilation does not hide in the numbers. Brian2 CPU is the accuracy reference: comparison scripts report active-neuron Jaccard overlap, per-neuron firing-rate correlation and spike-count ratios between backends. Spike exports are written outside the timed section, with a manifest and checksums.

Official site ↗

#benchmark#brian2#brian2cuda#pytorch#nest-gpu#genn#gpu