Comparison of semi-implicit and explicit finite difference algorithms on highly parallel processing architectures

Modern GPUs (graphical processing units) are a common source of processing power inmany supercomputers. Their performance derives from the highly parallel architecture that is em-ployed and have the benefit of low cost, temperature and power consumption. Two finite differencemodels have been implemented on GPU, a semi-implicit and an explicit algorithm, to numericallymodel a stratified shear layer, that needs fine meshes to be modelled accurately. The GPU modelswere shown to improve performance by factors of around 50x and 20x for the semi-implicit and ex-plicit models respectively