Computational Science Technical Note CSTN-070

Numerical Simulation of the Complex Ginzburg-Landau Equation on GPUs with CUDA

K.A.Hawick and D.P.Playne

Archived January 2009, Revised September 2010

Abstract

The Time Dependent Ginzburg Landau(TDGL) equation models a complex scalar field and is used to study a variety of different physical systems and exhibits phase transitional behaviours that necessitate study using numerical simulation methods. We employ fast data-parallel simulation algorithms on Graphical Processing Units (GPUs) and report on performance data and stability tradeoffs from using various implementations of both 32-bit and 64-bit complex numbers. Using NVIDIA's Compute Unified Device Architecture (CUDA) programming language running on a GTX480 GPU, we are able to simulate the TDGL with relatively large simulation system sizes of $256^3$ cells and we discuss the relative computational tradeoffs between numerical accuracy and stability using different methods as well as different data precisions.

Keywords: Ginzburg-Landau equation; PDE; complex numbers; numerical precision; CUDA, GPUs.

Full Document Text: PDF version.


[ CSTN Index ]