high performance computing on graphics processing units: hgpu.org

In the past years, graphic processing units have become a new abundant parallelcomputing resource on personal computers. In this work parallel computation ofa typical case in nite element analysis for solids has been practiced. The solutionof 3-D linear elastic static problems with 3 degree of freedom is fully implementedutilizing the current GPU technology. Discretization of the problem has been donefor two cases of: Tetrahedral and Hexahedral elements. Acceleration of the solutionhas been realized using SIMT parallel algorithms. Sparse matrix storage formats aswell as matrix-vector operations are investigated for optimum hardware utilization.preconditioned conjugate gradient method has been fully implemented on the GPUdevice as the iterative solver. FEA GPU implementation is compared with the correspondingoptimized serial version run on a conventional processor with the sametechnology for various mesh sizes, sparse matrix storage schemes, and choice of basis function.