Regensburg 2010 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe
MA: Fachverband Magnetismus
MA 33: Poster II
MA 33.31: Poster
Freitag, 26. März 2010, 11:00–14:00, Poster B1
Massively parallel micromagnetic FEM calculations with Graphical Processing Units — •Attila Kákay1, Elmar Westphal2, and Riccardo Hertel1 — 1Forschungszentrum Jülich GmbH, Institut für Festkörperforschung, Elektronische Eigenschaften — 2Forschungszentrum Jülich GmbH, IFF-Netzwerke und Numerik
We have adapted our Finite Element Micromagnetic simulation software to the massively parallel architecture of Graphical Processing Units (GPUs) by employing NVIDIA’s “Compute Unified Device Architecture” (CUDA)[1] software platform for high-performance computing. The performance of the code is demonstrated on the example of µMAG standard problem #4, where the computation time of a single GPU is compared with an OpenMP-parallelized version of our code using eight CPUs. In this comparison, the adaption of both the magnetostatic field calculation and the time integration of the Landau-Lifshitz-Gilbert equation to the GPU architecture can lead to a speedup factor of up to ten. The performance of the GPU code increases with increasing number of discretization points. The computation time required for high-resolution micromagnetic simulations of the magnetization dynamics in large magnetic samples can thus effectively be reduced by employing GPUs.
[1] CUDA Programming Guide, http://www.nvidia.com/