Parallel molecular dynamics for silicon and silicon carbide: MPI, CUDA and CUDA-MPI implementation

What is it about?

A parallel hybrid algorithm was developed and implemented as computer code in order to combine CUDA and MPI technologies in a single program. The efficiency of the created program was compared both with a program that uses only the MPI interface, and with a CUDA program designed to work only with one GPU. Despite the relative simplicity of making changes to the base MPI code, the resulting hybrid code demonstrated good performance and scaling. Сode examples were presented in the paper

Featured Image