The Link Cell Method Using Compute Unified Device Architecture and the Message Passing Interface Standard

This paper shows the parallel implementation of the Linked Cell Method using CUDA and MPI. The so called Linked Cell method is employed in particle simulations with short range interactions. The method divides the physical domain into uniform sub domains while limiting the interactions to particles occupying the same cell and the flanking ones. In CUDA the Graphic Processing Units (GPUs) can be used for general purpose calculations. In the present case, the MPI standard allows the distribution of the sub domains into the available GPUs. Achieving an optimum GPU interchange process is crucial for system scaling. Results here presented will be used in the operation of a multi-GPUs system for fluid dynamics simulations using Smoothed Particle Hydrodynamics (SPH).