Selke, Gunnar
[Author]
;
Möller, Dietmar P. F.
[Contributor]
Design and Development of a GPU-Accelerated Micromagnetic Simulator ; Entwurf und Entwicklung eines mikromagnetischen Simulators mit GPU-Beschleunigung
You can manage bookmarks using lists, please log in to your user account for this.
Media type:
Doctoral Thesis;
Electronic Thesis;
E-Book
Title:
Design and Development of a GPU-Accelerated Micromagnetic Simulator ; Entwurf und Entwicklung eines mikromagnetischen Simulators mit GPU-Beschleunigung
Contributor:
Selke, Gunnar
[Author]
Published:
Staats- und Universitätsbibliothek Hamburg Carl von Ossietzky, 2013-01-01
Footnote:
Diese Datenquelle enthält auch Bestandsnachweise, die nicht zu einem Volltext führen.
Description:
Micromagnetic simulators are important software tools to investigate ferromagnetic nano- and microstructures. In this work the fast micromagnetic simulator MicroMagnum is developed, which runs on CPUs as well as on graphics processing units (GPUs). Its high performance on GPUs allows the investigation of large simulation problems that could not be treated before by previous software. From the beginning MicroMagnum was developed with strong feedback by the users to obtain a high usability. Fast algorithms using the finite-difference method are presented that implement the micromagnetic model efficiently. By employing graphics processing units, a speedup of up to two orders of magnitude is achieved. Modern graphics processors allow general programming on graphic processing units. GPUs have many compute cores and their programming model is inherently parallel. Thus the algorithms have to be parallelized before they are ported to the GPU architecture. Several parallel software patterns are adapted to the GPU architecture and applied to the implementation of the micromagnetic model. The software requirements of performance, correctness, extendability, usability, portability, and maintainability are desired properties. The simulator is written in Python and C++. To achieve high software extendability, the dynamic scripting language Python is used for the implementation of the non-speed-critical software parts. Likewise the users of the simulator write their simulation scripts in Python. A module system that manages the dependencies of the model variables of the micromagnetic model is presented. An abstraction layer that hides the implementation details of the CPU and GPU algorithms and data representations is developed. This speed-critical software layer is written in C++ and CUDA C. Several benchmarks that compare the performance of the CPU and GPU routines are performed. As a use case, non-linear vortex-core dynamics are investigated using simulations performed by MicroMagnum. ; Mikromagnetische Simulatoren sind ...