A Comparison of Methods for Calculation of Transformation Matrices for Model Order Reduction Using GPU Parallel Computing

The purpose of this paper is to compare execution times for developed programs for determining transformation matrices in model order reduction using the balanced realization method. Six popular methods are implemented, named the: RPR, SR, BFSR, EIG-SR, EIGBFSR and Obinata-Anderson method. Each algorithm is prepared in two versions, one for execution on the main processor (CPU), and the other one for execution on graphics processor (GPU).