Towards efficient GPGPU Cellular Automata model implementation using persistent active cells

Abstract

Natural complex phenomena simulation relies on the application of advanced numerical models. Nevertheless, due to their inherent temporal and spatial computational complexity, efficient parallel computing algorithms are required in order to speed up simulation execution times. In this paper, we apply the Nvidia CUDA architecture to the simulation of a groundwater hydrological model based on the Cellular Automata formalism. Different implementations, using different memory access patterns and optimizations, regarding the application of persistent active cells (i.e., once a cell is activated, it remains such throughout a simulation), are presented and evaluated. The obtained results have demonstrated the full suitability of the approach in speeding up simulation times, thus resulting in a valid support for complex system modeling.