Title: GAME: GPU accelerated multipurpose evolutionary algorithm library

Authors: Péter Cserti; Szabolcs Szondi; Balázs Gaál; István Vassányi

Addresses: Department of Electrical Engineering and Information Systems, University of Pannonia, 8200 Veszprém, Egyetem út 10, Veszprem, Hungary ' Department of Electrical Engineering and Information Systems, University of Pannonia, 8200 Veszprém, Egyetem út 10, Veszprem, Hungary ' Department of Electrical Engineering and Information Systems, University of Pannonia, 8200 Veszprém, Egyetem út 10, Veszprem, Hungary ' Department of Electrical Engineering and Information Systems, University of Pannonia, 8200 Veszprém, Egyetem út 10, Veszprem, Hungary

Abstract: The use of genetic algorithms (GAs) has grown to widespread acceptance by providing an efficient way to solve complex problems lacking deterministic solvers. GAs employ a special stochastic search method based on evolutionary theory, which gives them the ability to outperform most traditional search algorithms. Also their use of independent individuals makes them an ideal candidate for parallelisation enhancing their inherently good performance even further. Their parallelisability on graphical processing units (GPU) had been shown multiple times, but the implementations were either single-objective GAs or just partially accelerated by GPUs, also every time they were experimental designs. The genetic algorithm library discussed in this article is the first that contains fully parallelised GPU implementations of multi-objective genetic algorithms besides the single-objective ones. Furthermore, it is organised into a ready to use framework, which provides flexible and efficient GPU accelerated GAs. Thus, enabling the user to solve complex problems faster than standard CPU-based implementations would allow and with lower overall energy cost.

Keywords: C/C++; genetic algorithms; graphical processing unit; GPU; many-threaded differential evolution; multi-objective GAs; Nvidia CUDA; parallelisation; parallel computing.

DOI: 10.1504/IJICA.2013.055936

International Journal of Innovative Computing and Applications, 2013 Vol.5 No.3, pp.163 - 172

Received: 01 Oct 2012
Accepted: 28 Nov 2012
Published online: 31 Jul 2014 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article