Parallel Implementation of the Heisenberg Model Using Monte Carlo on GPGPU
Universidade Federal de Juiz de Fora, DCC/ICE and DF/ICE, Cidade Universitaria, CEP: 36036-330, Juiz de Fora, MG, Brasil
In Computational Science and Its Applications – ICCSA 2011, Vol. 6784 (2011), pp. 654-667
@article{campos2011parallel,
title={Parallel Implementation of the Heisenberg Model Using Monte Carlo on GPGPU},
author={Campos, A. and Pe{c{c}}anha, J. and Pampanelli, P. and de Almeida, R. and Lobosco, M. and Vieira, M. and de O. Dantas, S.},
journal={Computational Science and Its Applications-ICCSA 2011},
pages={654–667},
year={2011},
publisher={Springer}
}
The study of magnetic phenomena in nanometer scale is essential for development of new technologies and materials. It also leads to a better understanding of magnetic properties of matter. An approach to the study of magnetic phenomena is the use of a physical model and its computational simulation. For this purpose, in previous works we have developed a program that simulates the interaction of spins in three-dimensional structures formed by atoms with magnetic properties using the Heisenberg model with long range interaction. However, there is inherent high complexity in implementing the numerical solution of this physical model, mainly due to the number of elements present in the simulated structure. This complexity leads us to develop a parallel version of our simulator using General-purpose GPUs (GPGPUs). This work describes the techniques used in the parallel implementation of our simulator as well as evaluates its performance. Our experimental results showed that the parallelization was very effective in improving the simulator performance, yielding speedups up to 166.
July 22, 2011 by hgpu
