Investigation of parallel computation in 3D magnetic field analysis on distributed memory type of multiprocessors containing 296 PUs Available to Purchase

The algorithm of parallelized conjugate gradient method for solving large‐scale linear equations is investigated. The performance of the algorithm is evaluated by solving sparse and full matrices on distributed memory type of parallel computer containing 296PUs. It is shown that the speed‐up ratio of the parallel computation of sparse matrix is saturated at several ten PUs due to large broadcasting time of data among PUs.