Title :
5.6: GPU enhancement of FDTD-PIC plasma-wave simulations
Author :
Woods, Andrew J. ; Ludeking, Lars D.
Author_Institution :
Alliant Techsystems (ATK), Newington, VA, USA
Abstract :
Simple models of major CPU-intensive MAGIC electromagnetic (EM) plasma code portions using the CUDA language run on the graphical processing unit (GPU) indicate 12x computing rate compared to the same calculations run on the CPU only. MAGIC is being modified for performance speedup of large-scale plasma-wave EM calculations using GPU processing. Results to-date from MAGIC with the particle update segment converted to the GPU show no net gain in execution speed. Net gain will be realized when CPU to GPU transfer time is further offset by converting larger portions of the code to the GPU.
Keywords :
finite difference time-domain analysis; physics computing; plasma electromagnetic wave propagation; plasma simulation; CUDA language; FDTD-PIC; GPU enhancement; MAGIC; electromagnetic plasma code; graphical processing unit; plasma-wave simulations; Central Processing Unit; Computer architecture; Concurrent computing; Electromagnetic modeling; Large-scale systems; Parallel processing; Physics; Plasma applications; Plasma materials processing; Plasma simulation; CUDA; FDTD-PIC; GPU; MAGIC;
Conference_Titel :
Vacuum Electronics Conference (IVEC), 2010 IEEE International
Conference_Location :
Monterey, CA
Print_ISBN :
978-1-4244-7098-3
DOI :
10.1109/IVELEC.2010.5503600
