An efficient GPU implementation of a multi-start TSP solver for large problem instances

Authors:
Kamil Rocki;Reiji Suda
Affiliations:
The University of Tokyo, CREST, JST, Tokyo, Japan;The University of Tokyo, CREST, JST, Tokyo, Japan
Venue:
Proceedings of the 14th annual conference companion on Genetic and evolutionary computation
Year:
2012

Citing 2
Cited 0

The Traveling Salesman Problem: A Computational Study (Princeton Series in Applied Mathematics)

The Traveling Salesman Problem: A Computational Study (Princeton Series in Applied Mathematics)
A highly-parallel TSP solver for a GPU computing platform

NMA'10 Proceedings of the 7th international conference on Numerical methods and applications

Quantified Score

Hi-index	0.00

Visualization

Abstract

We are presenting a parallel GPU implementation of the Traveling Salesman Problem (TSP) solver. Our method is based on the iterative hill climbing algorithm first proposed by O'Neil et al., but modified in order to solve large instances of the problem. Our results show that GPU can be used to find an approximate solution of a problem which size is up to 6 thousand cities in a very efficient way achieving over 170 GFLOPS on a single TESLA C2050 card running thousands of independent threads. Due to computation-oriented architecture of GPUs, a more universal algorithm can be obtained with relatively little performance degradation.