high performance computing on graphics processing units: hgpu.org

Several acoustic simulation methods have been introduced during the past decades. Wave-based simulation methods have been one of the alternatives, but their applicability for wideband acoustic simulation has been limited by the computing power of available hardware. During recent years, the processing power and programmability of graphics processing units have improved, and therefore several wave-based simulation methods have become potential alternatives. In this thesis, a finite-difference time-domain solver is implemented. The performance of the solver is accelerated with the use of graphics processing units. Different performance considerations are reviewed and the system is evaluated by comparing the simulated responses to known analytic solutions. The resulting system is C++ software, which is interfaced with Matlab with the use of a mex-function. It is found that the forward difference boundary formulation is the most efficient for parallel implementation due to a lesser number of operations. The usage of double precision data type in the simulation decreases the performance significantly. The system is found to follow the analytical solutions with accuracy expected of the method, apart from the reflection characteristics of the forward difference boundary formulation that deviate slightly from the analytical solution.