Speeding up an observing system simulation experiment with GPU in a Matlab framework

CINTIA BRUSCANTINI

Mendoza

Simposio; Vi Latin American Symposium on High Performance Computing; 2013

Universidad Nacional de Cuyo

Space agencies make use of simulations of Earth-observing satellite system in order to perform the mission design and analysis of the characteristics of the final product for the users. We developed a simulation for the Aquarius/SAC-D mission for assessing the accuracy of soil moisture retrieval from passive and active L-band remote sensing. The simulation was implemented in a Matlab-based framework and is currently running in a single core. However, Matlab offers several alternatives to speed up the computation. After performing a time profile of the entire simulation and analyzing part of the code that can be parallelized, a time comparison can be carried out when running the code on a single core (eg. for), multiple cores (eg. parfor), using gpu from a single core (eg. gpuArray) and from multiple cores (eg. gpuArray and parfor).