Parallelizing a Micromagnetic Program for Use on Multiprocessor Shared Memory Computers

Parallelization of a finite difference micromagnetic program on shared memory computer systems is studied. Efficiency is found to be limited by memory bandwidth, and techniques are introduced to reduce memory traffic. Computations are sped up by a factor of three with four processor cores; a factor of five is possible on some systems. This corresponds to a Karp-Flatt serial fraction of 5-10% for small core counts.