Effective use of programmable graphics processing units in problems of molecular dynamics simulation

Graphics processing units (GPU) make it possible to expand significantly the capabilities of computing systems. The article discusses the use of graphics processors in problems of molecular dynamics modeling with a complex particle interaction potential. In order to improve the performance of calculations on the GPU, the following methods are implemented: reduction of the number of requests to the global memory; reduction of the number of branches; selection of the optimal load of multiprocessors; and use of equivalent mathematical expressions that perform functions for faster code execution. Parallel execution of the program is realized by partitioning the space into cells, compiling and updating the list of neighboring atoms in order to minimize memory collisions, distribution of operations over computational flows, and allocation of additional memory for creating copies of the coordinates of interacting atoms. Several examples illustrating the computational complexity of the algorithms are considered.