Acceleration of electromagnetic launchers modeling by using graphic processing units

The solution of large and complex coupled electromechanical problems requires high performance computing resources. In the last years the use of Graphic Processing Units (GPUs) has gained increasing popularity in scientific computing because of their low cost and parallel architecture. In this paper the authors report the main results of the GPU approach to the parallelization of a research code for the electromagnetic launcher analysis. Programming a GPU - based environment poses a number of critical issues that have to be carefully addressed in order to fully exploit the potentiality of the system. Data have to be properly organized in order to fit the Single Instruction Multiple Data scheme; the data transfer between the host and the device, as well as the memory management of the GPU deserve accurate programming. Two examples of application of the parallelized code have been reported to show the performance improvements that can be obtained in the numerical analysis of both rail and induction launchers.