Some comments about graphic processing unit (GPU) architectures applied to finite-difference time-domain (FDTD) room acoustics simulation: Present and future trends

• Published in: Proceedings of Meetings on Acoustics Vol. 19; no. 1
• Main Authors: Lopez, Jose J., Navarro, Juan M., Carnicero, Diego, Escolano, Jose
• Format: Conference Proceeding
• Language: English
• Published: 02.06.2013
• ISSN: 1939-800X
• DOI: 10.1121/1.4800576

The parallelization of the finite-difference time-domain (FDTD) method for room acoustic simulation using graphic processing units (GPUs) has been subject of study even prior to the introduction of general-purpose computing environments such as the CUDA architecture. Nowadays CUDA offers enough flexibility and processing power to obtain performance gains higher than 200 times compared to single-threaded CPU codes. In this paper, different aspects related to the implementation of FDTD in CUDA are analyzed; first, how the evolution of the different CUDA architectures affects implementations is inquired, paying special attention to the Kepler architecture, the latest available. Also performance increasing by using the different memory subsystems the GPU offers is discussed. Moreover, the performance in the use of the available computing power in the GPU is also analyzed together with the limiting factors such as memory consumption and computing time that prevent the simulation of large rooms at very high frequencies. Next the hybrid methods that combine FDTD with other less expensive room acoustic methods are commented. Finally the application of GPU for the acceleration of the Diffusion Equation Method for room acoustic simulation is discussed.