The accelerated multi-stage depressed collector code COLLGUN using CUDA

Abstract

COLLGUN, designed to simulate multistage depressed collectors, was optimized by using parallel computing. This work reports the speedups observed on a parallel implementation of the COLLGUN code using CUDA. Were parallelized the solver of the sparse system for the electric scalar potential using the Incomplete Cholesky Conjugate Gradient Method (a) and the determination of the electric field inside each tetrahedron of the mesh (b), and speedups over 3X and 5X were respectively observed.