Speaker
Description
This master's thesis focuses on extending the accelerated implementation of acoustic wave propagation simulation in a medium using the k-Wave toolbox to enable the use of hybrid local decomposition on multi-GPU systems. The aim of this work is to implement a framework that allows decomposing the simulation domain into subdomains and computing the simulation with different temporal and spatial resolutions for individual subdomains. This thesis explains the method of simulating sound propagation and the technologies used in programming for multi-GPU systems. Furthermore, the thesis presents design of the implementation and describes the implementation itself. Finally, it summarizes the results achieved. A maximum error of 0.8% and a 2.8-fold speedup were achieved when decomposing into 8 subdomains without change in resolution. 1.93-fold speedup was achieved with change in resolution.
