Conjugate gradient algorithm for solving linear systems with block-fivediagonal matrices on GPU

摘要

For solving systems of linear equations with the block-fivediagonal matrices, the parallel algorithms on the basis of the iterative conjugate gradient method with regularization and the conjugate gradient method with preconditioner are developed and implemented numerically on graphics processors using the CUDA technology. The comparison of computation times of parallel algorithms for the Intel processor and NVIDIA Tesla GPU is carried out. The problem with quasi-model data is solved.