Parallel Genetic Algorithm Solving 0/1 Knapsack Problem Running on the GPU
Brno University of Technology Faculty of Information Technology, Department of Computer Systems, Bozetechova 2, 612 66 Brno, Czech Republic
16th International Conference on Soft Computing MENDEL 2010, p.64-70
@inproceedings{Pospichal9253,
author={Petr Pospichal and Josef Schwarz and Jiri Jaros},
title={Parallel Genetic Algorithm Solving 0/1 Knapsack Problem Running on the GPU},
pages={64–70},
booktitle={16th International Conference on Soft Computing MENDEL 2010},
year={2010},
location={Brno, CZ},
publisher={Brno University of Technology},
ISBN={978-80-214-4120-0},
language={english},
url={http://www.fit.vutbr.cz/research/view_pub.php?id=9253}
}
In this work, we show that consumer-level $100 GPU can be used to significantly speed-up optimization of 0/1 Knapsack problem. We identify strong and weak points of GPU architecture and propose our parallel genetic algorithm model implemented in CUDA running entirely on the GPU. We show that GPU must be utilized for sufficiently long time in order to obtain reasonable program speedup. Then we compare results quality and speed of our model with single-threaded CPU code implemented using Galib. Peak speedup of GPU GA execution performance is 1340x resp. 134x for 4-bit resp. 40-bit problem instances while maintaining reasonable results quality.
January 22, 2011 by hgpu
