Scalable GPU-Parallelized FDTD Method for Analysis of Large-Scale Electromagnetic Dosimetry Problems.

J Chakarothai, K Wake… - Applied Computational …, 2016 - search.ebscohost.com
Applied Computational Electromagnetics Society Journal, 2016search.ebscohost.com
A massively parallel finite-difference timedomain (FDTD) method using a GPU cluster of
TSUBAME system has been developed for numerical exposure of a human body to
electromagnetic fields. We have also developed and implemented a novel approach for
tracing total energy absorbed into a human body in our parallel FDTD code. Our developed
FDTD code has shown a strong scalability using 216 nodes and 648 GPUs in total with a
high capability to calculate over ten billion cells per second. Whole-body specific absorption …
Abstract
A massively parallel finite-difference timedomain (FDTD) method using a GPU cluster of TSUBAME system has been developed for numerical exposure of a human body to electromagnetic fields. We have also developed and implemented a novel approach for tracing total energy absorbed into a human body in our parallel FDTD code. Our developed FDTD code has shown a strong scalability using 216 nodes and 648 GPUs in total with a high capability to calculate over ten billion cells per second. Whole-body specific absorption rate (SAR) at 200 MHz, as well as its distribution, of a human body with a 0.5 mm resolution with 40000 time steps was found in less than approximately three hours, showing the availability of the method for large-scale EM dosimetry problems.
search.ebscohost.com
Showing the best result for this search. See all results