Detail publikace

GPU-Based Acceleration of the Genetic Algorithm

Originální název

GPU-Based Acceleration of the Genetic Algorithm

Anglický název

GPU-Based Acceleration of the Genetic Algorithm

Jazyk

en

Originální abstrakt

Genetic algorithm, a robust, stochastic optimization technique, is effective insolving many practical problems in science, engineering, and business domains. Unfortunatelly, execution usually takes a long time. In this paper, I study possibility of utilization consumer-level graphics cards for acceleration of GA's. A mapping of parallel island genetic algorithm to CUDA software model is designed and tested on GeForce 8800GTX, GTX260-SP216 and GTX285 GPU's using Rosenbrock's, Griewank's and Michalewicz's benchmark functions. Results indicates that this optimization leads to speedups up to seven thousand times compared to single CPU thread while maintaing reasonable results quality.

Anglický abstrakt

Genetic algorithm, a robust, stochastic optimization technique, is effective insolving many practical problems in science, engineering, and business domains. Unfortunatelly, execution usually takes a long time. In this paper, I study possibility of utilization consumer-level graphics cards for acceleration of GA's. A mapping of parallel island genetic algorithm to CUDA software model is designed and tested on GeForce 8800GTX, GTX260-SP216 and GTX285 GPU's using Rosenbrock's, Griewank's and Michalewicz's benchmark functions. Results indicates that this optimization leads to speedups up to seven thousand times compared to single CPU thread while maintaing reasonable results quality.

BibTex


@inproceedings{BUT34930,
  author="Petr {Pospíchal}",
  title="GPU-Based Acceleration of the Genetic Algorithm",
  annote="Genetic algorithm, a robust, stochastic optimization technique, is effective
insolving many practical problems in science, engineering, and business domains.
Unfortunatelly, execution usually takes a long time. In this paper, I study
possibility of utilization consumer-level graphics cards for acceleration of
GA's. A mapping of parallel island genetic algorithm to CUDA software model is
designed and tested on GeForce 8800GTX, GTX260-SP216 and GTX285 GPU's using
Rosenbrock's, Griewank's and Michalewicz's benchmark functions. Results indicates
that this optimization leads to speedups up to seven thousand times compared to
single CPU thread while maintaing reasonable results quality.",
  address="Faculty of Information Technology BUT",
  booktitle="Počítačové architektury a diagnostika 2010",
  chapter="34930",
  edition="NEUVEDEN",
  howpublished="print",
  institution="Faculty of Information Technology BUT",
  year="2010",
  month="september",
  pages="75--80",
  publisher="Faculty of Information Technology BUT",
  type="conference paper"
}