MPI for SPH Methods and Parallel Computing on CPUs and GPUs

---

Abstract

Message Passing Interface (MPI) is a standard designed for parallel programming on distributed memory systems, enabling multiple processors to work together by dividing and distributing tasks. This paper presents a comprehensive approach to addressing computational challenges in smoothed particle hydrodynamics (SPH) simulations through a novel MPI-based parallel SPH code. The research emphasizes code optimization for both CPU and GPU architectures, incorporating CUDA parallel programming to enhance GPU performance. Detailed insights into the code design, implementation, algorithm flowchart, and multi-GPU usage with MPI are provided. Experimental results demonstrate the model’s efficiency and scalability across various scenarios, laying a solid foundation for advancing research in fluid dynamics and parallel computing.

---

Keywords

Message Passing Interface (MPI); Smoothed Particle Hydrodynamics (SPH); GPU; Parallel Computing; CUDA