An Extensible, Data-Oriented Architecture for High-Performance, Many-World Simulation

https://madrona-engine.github.io/

Madrona batch ECS engine. Project page: Hide and Seek over 1.9M steps/s, Overcooked-AI 40M steps/s, Hanabi 20M steps/s (RTX 4090); 2-3 orders of magnitude over open-source CPU baselines, 5-33x over 32-thread CPU baselines.

Massively parallel GPU simulation