GPU-Accelerated Parallel FDTD on Distributed Heterogeneous Platform

Table 2

Statistics of the time consumption of cudaFDTD.

Function module	Proportion
𝐄 update	48.2%
Inter- and intraprocess communications of 𝐄	5.0%
Collection and distribution boundary data of 𝐄	2.2%
Source excitation	0.6%
𝐇 update	35.9%
Inter- and intraprocess communications of 𝐇	3.6%
Collection and distribution boundary data of 𝐇	3.8%
near-to-far transformation	0.7%