GPU-Based Simulation of Evolutionary Spatial Cyclic Games: A Comparative Evaluation of Apple Silicon vs Nvidia
Louie Sinadjan, Dave Cliff · Procedia Computer Science · 2025
We present the design, implementation and evaluation of GPU-accelerated simulators for Evolutionary Spatial Cyclic Games (ESCGs), a class of minimal agent-based models used to study the co-evolutionary dynamics of biodiversity in ecosystems. Traditional single-threaded ESCG simulations are computationally expensive and scale poorly. Contemporary many-core GPUs offer the possibility of greatly reducing ESCG simulation runtimes via multi-threaded parallel implementations, but (as far as we are aware) there are no prior publications describing GPU implementations of ESCGs, and thus the novel contribution of this paper is that we believe it to be the first to describe and rigorously evaluate the performance of ESCGs implemented on GPUs. To address this, we have developed high-performance ESCG implementations using Apple’s Metal Shading Language (MSL) and Nvidia’s CUDA, and have also independently developed single-threaded versions in C++ and C for cross-validation, from which we generated baseline performance measures. Benchmarking results show that GPU acceleration delivers significant speedups, with our CUDA implementation achieving at best a speed-up of roughly 28x. We tested ESCG system sizes up to 3200x3200 and found that our CUDA implementation remains tractable while MSL faced scalability issues. Our code is available on GitHub, freely available under the MIT Open Source License, for other developers to adapt and extend, and as a platform for other ESCG researchers to use in their work.