A Comparative Study of MAGMA and cuBLAS Libraries for GPU based Vector Fitting

System identification from tabulated data has gained significant attention in the recent years for signal and power integrity analysis of high-speed circuits and interconnects. Vector Fitting (VF) algorithm has been widely used for this purpose. Recently GPU based Vector Fitting (GVF) was proposed to exploit the massive parallel processing capabilities of the emerging GPU platforms. In this paper, a detailed comparative performance study of MAGMA and cuBLAS libraries while implementing GVF is presented. Results demonstrate that use of MAGMA libraries provide significantly better performance compared to cuBLAS libraries.

[1]  B. Gustavsen,et al.  Improving the pole relocating properties of vector fitting , 2006, 2006 IEEE Power Engineering Society General Meeting.

[2]  S. Grivet-Talocia,et al.  A parallel Vector Fitting implementation for fast macromodeling of highly complex interconnects , 2010, 19th Topical Meeting on Electrical Performance of Electronic Packaging and Systems.

[3]  S. Grivet-Talocia,et al.  On the Parallelization of Vector Fitting Algorithms , 2011, IEEE Transactions on Components, Packaging and Manufacturing Technology.

[4]  B. Gustavsen,et al.  Computer Code for Rational Approximation of Frequency Dependent Admittance Matrices , 2002, IEEE Power Engineering Review.

[5]  T. Dhaene,et al.  Macromodeling of Multiport Systems Using a Fast Implementation of the Vector Fitting Method , 2008, IEEE Microwave and Wireless Components Letters.

[6]  A. Semlyen,et al.  Rational approximation of frequency domain responses by vector fitting , 1999 .