Table of Content

Open Access

ARTICLE

Local strong form meshless method on multiple Graphics Processing Units

G. Kosec1,2, P. Zinterhof3
Jožef Stefan Institute, Department of Communication Systems, Jamova 39 1000 Ljubljana, Slovenia.Email: gkosec@ijs.si
University of Nova Gorica, Laboratoty for Multiphase processes, Vipavska 13, 5000 Nova Gorica, Slovenia.
University of Salzburg, Jakob-Haringer-Str. 2, 5020 Salzburg, Austria.
Email:peter.zinterhof3@sbg.ac.at

Computer Modeling in Engineering & Sciences 2013, 91(5), 377-396. https://doi.org/10.3970/cmes.2013.091.377

Abstract

This paper deals with the implementation of the local meshless numerical method (LMM) on general purpose graphics processing units (GPU) in solving partial differential equations (PDE). The local meshless solution procedure is formulated in a way suitable for parallel execution and has been implemented on multiple GPUs. The implementation is tested on a solution of diffusion equation in a 2D domain. Different setups of the meshless approach regarding the selection of basis functions are tested on an interval up to 2.5 million of computational points. It is shown that monomials are a good selection of the basis when working with a high number of nodes. The results are presented in terms of error analysis, convergence analysis and computational performance measurements.

Keywords

Local Meshless Method, LRBCFM, DAM, Dirichlet jump, GPU, CUDA, dynamic load balancing

Cite This Article

Kosec, G., Zinterhof, P. (2013). Local strong form meshless method on multiple Graphics Processing Units. CMES-Computer Modeling in Engineering & Sciences, 91(5), 377–396.



This work is licensed under a Creative Commons Attribution 4.0 International License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
  • 750

    View

  • 543

    Download

  • 0

    Like

Share Link

WeChat scan