A comparative study of massive flock simulations on the cpu and the gpu

Abstract

This paper describes how large scale flocks can be simulated efficiently with modern computer graphics using parallelization techniques and spatial data structures. The results show that parallelism is essential in improving performance, both in CPU and GPU implementations. Best performance was achieved using CUDA, uniform grid, and batching which resulted in a simulation of two million agents at a frame rate of 30 frames per second. The simulation was extended with a graphical user interface, predators, colours and user interaction in order to create a simple game.