Genetic algorithm, a robust, stochastic optimization technique, is effective in solving many practical problems in science, engineering, and business domains. Unfortunatelly, execution usually takes long time. In this paper, we study a possibility of utilization consumer-level graphics cards for acceleration of GAs. We have designed a mapping of the parallel island genetic algorithm to the CUDA software model and tested our implementation on GeForce 8800GTX and GTX285 GPUs using a Rosenbrock's, Griewank's and Michalewicz's benchmark functions. Results indicates that our optimization leads to speedups up to seven thousand times compared to single CPU thread while maintaing reasonable results quality. |