379 Performance Optimizations for TTI RTM on GPU based Hybrid Architectures

There has been an ever increasing gap between energy demand and supply of oil and gas worldwide due to multiple challenges including low recovery rates from existing wells, complications of deepsea exploration and others. This has led to exploration of unconventional hydrocarbon resources including Shale Oil & Gas and others. These factors worldwide have increased the dependence of the petroleum industry on research in High Performance Computing to provide faster and more accurate solutions for simulation, modelling and prediction of Oil & Gas resources. Seismic Imaging plays a major role in providing estimates of resources for new reservoirs as well as for existing reservoirs especially when used for history matching and reservoir characterization. Reverse Time Migration is a state of the art technique used in Seismic Imaging and Full Waveform Inversion and is being used widely for complex sub-salt imaging. However, it has huge computational cost which makes it challenging for large-scale exploration. In this paper, we present performance optimizations for TTI RTM algorithm on hybrid GPU based architectures and demonstrate around 4× end-end performance gain over CPU only runs. This demonstration of endend performance gain for a production code is a unique contribution as compared to the prior work.