1. Introduction

The model size increases, and calculation time is required for highly accurate calculations, such as models consisting of a large number of complex components close to the actual machine, or for detailed loss density distributions. Under such situations, the computation time can be reduced by using parallel processing. Three parallel calculation options are available for magnetic field analysis. Specifically, these are CPU-based shared memory type (SMP), massive parallel distributed memory type (MPP), and parallel computing using a graphic processing unit (GPU). MPPs that show a high speed increase are effective when the model scale is large, from hundreds of thousands of elements to millions and so on.
This datasheet presents the computational speed-up performance of the MPP solver with respect to the number of parallels. MPP solvers show a near linear speed-up ratio with respect to the number of parallels if the number of elements is high. Generally, it is difficult to obtain a speed-up ratio when the number of elements is less than one million, but JMAG’s MPP solver maintains a high speed-up ratio up to 512 parallel elements. Computation times are also presented based on a large-scale calculation example: a model with 3 million elements can be computed in a few hours.