Implementing the Global Cellular Automata on CUDA

Abstract

The Global Cellular Automata (GCA) is a generalization of the Cellular Automata. As a massively parallel model, it is used to describe complex systems and algorithms in a coherent way. In this work, we evaluated how the generic GCA model transfers to NVIDIA's CUDA architecture on GPUs using two exemplary GCA algorithms. We compared our CUDA implementations with parallel CPU implementations and a fast and optimized 32-pipeline FPGA implementation. We obtained more than one order of magnitude in performance gain compared to standard CPUs, while our system also compared favourably to the FPGA implementations, showing that the GCA model fits well to current graphics cards using CUDA.