CUDA Based Fast Implementation of Very Large Matrix Computation