(RP07) GPU-accelerated Simulation of Elastic Wave Propagation
Performance Analysis and Optimization
TimeTuesday, June 26th8:30am - 10am
LocationSubstanz 1, 2
DescriptionModeling of ultrasound waves propagation in hard biological materials such as bones and the skull has a rapidly growing area of applications, e.g., brain cancer treatment planning, deep brain neurostimulation and neuromodulation, and opening blood-brain barriers. Recently, we have developed a novel numerical model of elastic wave propagation based on the Kelvin-Voigt model accounting for linear elastic wave proration in heterogeneous absorption media. Although the model offers unprecedented fidelity, its computational requirements have been prohibitive for realistic simulations. This paper presents an optimized version of the simulation model accelerated by the Nvidia CUDA language and deployed on the best GPUs including the Nvidia P100 accelerators in the Piz Daint supercomputer. The native CUDA code reaches a speed-up of 5.4 when compared to the Matlab prototype accelerated by the Parallel Computing Toolbox running on the same GPU. Such reduction in computation time and exclusion of Matlab from the critical path, enables to compute complex treatment plans in terms of hours.