Finite element (FE) analyses of large ordered systems require the solution of a large set of simultaneous coupled equations. The solution of these equations can be computationally expensive. This is especially true if one is carrying out stochastic analysis of large ordered systems where either the material parameters or the loading or both are assumed to be stochastic. A traditional approach that does not take into account any approximating methods requires taking recourse to Monte Carlo simulations. This involves repeated deterministic solutions of the FE problem corresponding to different realizations. This increases the computational cost further. The focus of this study is to investigate the use of Graphical Processing Unit (GPU) in parallel computations so as to reduce the computational costs. Numerical codes have been developed that interface between commercially available FE software, MATLAB and GPU enabled algorithms. The developments have been demonstrated using two numerical examples. The first problem considers the nonlinear response analysis of a 2D plane stress plate with random spatial inhomogeneities. The second problem considers the dynamic analysis of cantilever beam with a tip concentrated load. A comparison of the CPU times between the traditional sequential programming and GPU enabled programs has been presented.

Keywords: Parallel computing, Monte carlo simulation, Graphic processing unit.