Implementing genetic algorithms to CUDA environment using data parallelization

Računarske metode rješavanja paralelnih problema korištenjem grafičkih obradnih jedinica (GPUs) zadnjih su godina pobudile veliki interes. Paralelno izračunavanje može se primijeniti na genetske algoritme (GAs) u odnosu na proces evaluacije jedinki u populaciji. Ovaj rad opisuje još jednu metodu primjene GAs na CUDA okruženje gdje je CUDA računarsko okruženje opće namjene za GPUs koje daje NVIDIA. Osnovna karakteristika ovog istraživanja leži u tome da se paralelna obrada koristi ne samo za jedinke nego i za gene u jedinki. Predložena implementacija se procjenjuje kroz osam ispitnih funkcija. Ustanovili smo da predložena metoda implementacije daje 7,6-18,4 puta brže rezultate od onih kod primjene CPU.Computation methods of parallel problem solving using graphic processing units (GPUs) have attracted much research interests in recent years. Parallel computation can be applied to genetic algorithms (GAs) in terms of the evaluation process of individuals in a population. This paper describes yet another implementation method of GAs to the CUDA environment where CUDA is a general-purpose computation environment for GPUs provided by NVIDIA. The major characteristic point of this study is that the parallel processing is adopted not only for individuals but also for the genes in an individual. The proposed implementation is evaluated through eight test functions. We found that the proposed implementation method yields 7,6-18,4 times faster results than those of a CPU implementation