Efficient Parallel Preconditioned Conjugate Gradient Solver on GPU for FE Modeling of Electromagnetic Fields in Highly Dissipative Media

Author(s)

Peixoto de Camargos, Ana Flavia ^{[1, 2]} ; Silva, Viviane Cristine ^[1] ; Guichon, Jean-Michel ^[3] ; Munier, Gerard ^[3]

Affiliation

^[1] Univ Sao Paulo, Escola Politecn, BR-05508010 Sao Paulo - Brazil

^[2] Inst Fed Educ Ciencia & Tecnol Minas Gerais, Dept Engn, BR-35570000 Formiga, MG - Brazil

^[3] Ecole Natl Super Electrochim & Electrome Grenoble, Thermodynam & Physicochim Met Lab, CNRS, Lab Genie Elect Grenoble, F-38402 St Martin Dheres - France

• Document type: Journal article
• Source: IEEE Transactions on Magnetics; v. 50, n. 2 FEB 2014.
• Web of Science Citations: 7

Abstract

We present a performance analysis of a parallel implementation of preconditioned conjugate gradient solvers using graphic processing units with compute unified device architecture programming model. The solvers were optimized for the solution of sparse systems of equations arising from finite-element analysis of electromagnetic phenomena involved in the diffusion of underground currents in both steady state and under time-harmonic current excitation. We used both shifted incomplete Cholesky factorization and incomplete LU factorization as preconditioners. The results show a significant speedup using the graphics processing unit compared with a serial CPU implementation. (AU)

• FAPESP's process: 13/08201-5 - Compumag 2013
• Grantee: Viviane Cristine Silva
• Support Opportunities: Research Grants - Meeting - Abroad