Implementation of the FDTD method block algorithm in the MATLAB language using a graphics processing unit. 2D-decomposition case

Abstract

Abstract The problem of limited video memory when organizing parallel computing using the FDTD method on a non-professional graphics processor was considered in this article. As a solution, a block algorithm of the FDTD method with 2-D decomposition and its implementation in the MATLAB language are proposed. As a result, the problem of limited graphics memory was solved, the maximum possible discretization of the grid in calculations on the GPU was expanded from 10 million to 85 million nodes with an average acceleration of 7,5 times for the two-dimensional case of the FDTD method.