GPU-based 3D-FDTD computation for electromagnetic field dosimetry
Tomoaki Nagaoka, Soichi Watanabe
Abstract
Tomoaki Nagaoka, Soichi Watanabe
Abstract
Numerical dosimetry with the computational human model using the finite-difference time-domain (FDTD) method has recently been used for a safety assessment of electromagnetic field applications. However, the FDTD calculation runs very slowly and requires a large amount of computational memory. We focus, therefore, on general purpose programming on the graphics processing unit (GPGPU). We implemented the three-dimensional (3D) FDTD method on GPUs using Compute Unified Device Architecture (CUDA). In this study, we used the NVIDIA Tesla C2070 as a GPGPU board and tested the performance of FDTD computation on GPUs. The results indicated that while the single GPU/CPU speed ratio varies depending on the calculation domain, 3D-FDTD computation using a GPU requires significantly less run time than that using a conventional CPU. We confirmed that the FDTD computation on a multi-GPU is much faster than that on a single GPU, and we also found that eight GPUs can compute faster than a vector supercomputer.
OpenAlex reports 7 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Numerical dosimetry with the computational human model using the finite-difference time-domain (FDTD) method has recently been used for a safety assessment of electromagnetic field applications. However, the FDTD calculation runs very slowly and requires a large amount of computational memory. We focus, therefore, on general purpose programming on the graphics processing unit (GPGPU). We implemented the three-dimensional (3D) FDTD method on GPUs using Compute Unified Device Architecture (CUDA). In this study, we used the NVIDIA Tesla C2070 as a GPGPU board and tested the performance of FDTD computation on GPUs. The results indicated that while the single GPU/CPU speed ratio varies depending on the calculation domain, 3D-FDTD computation using a GPU requires significantly less run time than that using a conventional CPU. We confirmed that the FDTD computation on a multi-GPU is much faster than that on a single GPU, and we also found that eight GPUs can compute faster than a vector supercomputer.
Key concepts: Finite-difference time-domain method, CUDA, Computational science, Graphics processing unit, Computer science, General-purpose computing on graphics processing units, Parallel computing, Computation