Share Email Print

Proceedings Paper

CUDA-MPI-FDTD implementation of Maxwell's equations in general dispersive media
Author(s): Mohammad R. Zunoubi; Jason Payne; William P. Roach
Format Member Price Non-Member Price
PDF $17.00 $21.00

Paper Abstract

In this research, we present the first MPI-CUDA implementation of Finite-Difference Time-Domain (FDTD) discretization of Maxwell's equations in dispersive media that uses the MPI API to assign each CPU node its share of the computational domain and GPUs to their corresponding CPU threads. By taking advantage of the CUDA programming model, we present a unique implementation of the FDTD scheme that exploits the memory hierarchy of GPUs, including the global, texture, and shared memory. This enables us to tackle problems that are otherwise computationally prohibitive. Practical results will be presented along with a measure of speedup factors achieved when using multiple GPU processors.

Paper Details

Date Published: 9 February 2012
PDF: 6 pages
Proc. SPIE 8221, Optical Interactions with Tissue and Cells XXIII, 822115 (9 February 2012); doi: 10.1117/12.918429
Show Author Affiliations
Mohammad R. Zunoubi, Air Force Research Lab. (United States)
Jason Payne, Air Force Research Lab. (United States)
William P. Roach, Air Force Research Lab. (United States)

Published in SPIE Proceedings Vol. 8221:
Optical Interactions with Tissue and Cells XXIII
E. Duco Jansen; Robert J. Thomas, Editor(s)

© SPIE. Terms of Use
Back to Top
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research
Forgot your username?