The directions of computational efficiency scaling in numerical solution of typical element of wide range of different mathematical physics problems statements and also one of the most demanding to computer resources consumption – the Poisson equation have been considered. Special attention is given to studying the common efficiency of CUDA and MPI technologies of algorithms parallelization. It has been shown that the efficiency of MPI for solving the Poisson equation on computing clusters with NVIDIA graphic accelerators can be significantly improved as a result of increasing the dimension of the problem and for 3D case the nearly linear acceleration scaling of computations can be reached as a result of increasing the number of applied graphic accelerators.
parallel multithreaded and distributed computations, CUDA, MPI, Poisson equation
"Efektyvnist paralelnoho rozviazannia rivniannia Puassona na klasternykh obchysliuvalnykh systemakh z hrafichnymy pryskoriuvachamy" [Efficiency of parallel solving the Poisson equation on cluster computer systems with graphic accelerators],
Scientific Works of Kharkiv National Air Force University,