In order to design cellular automata rules by means of evolutionary algorithms, high computational demands need to be met. This problem may be partially solved by parallelization. Since parallel supercomputers and server clusters are expensive and often overburdened, this paper proposes the evolution of cellular automata rules on small and inexpensive graphic processing units. The main objective of this paper is not to evolve any actual cellular automata but to demonstrate that evolution of cellular automata rules can be accelerated significantly using graphics processing units. Several methods of speeding-up the evolution of cellular automata rules are proposed, evaluated and compared, some with very good results.

In order to design cellular automata rules by means of evolutionary algorithms, high computational demands need to be met. This problem may be partially solved by parallelization. Since parallel supercomputers and server clusters are expensive and often overburdened, this paper proposes the evolution of cellular automata rules on small and inexpensive graphic processing units. The main objective of this paper is not to evolve any actual cellular automata but to demonstrate that evolution of cellular automata rules can be accelerated significantly using graphics processing units. Several methods of speeding-up the evolution of cellular automata rules are proposed, evaluated and compared, some with very good results. (en)

cellular automata, parallel computing, GPU, CUDA, genetic alghorithm (en)