Third EAGE Workshop on HPC in Americas

Reverse Time Migration (RTM) is a preferred solution for modeling and imaging seismic data in areas of complex wave phenomena. It computes numerical solutions to the complete wave equation. Historically, it was impractical due to high computational costs. Nevertheless, developers have been porting their simulations to the new High-Performance Computing architectures, providing faster and more accurate results with each new generation. That makes RTM coupled with more precise velocity model-building workflows, now making RTM a viable option in the imaging portfolio. However, several challenges arrive when achieving high performance on HPC architectures. The first one is related to choosing the architecture that best fits your simulation. Moreover, the API used to implement the simulation affects the performance and portability of the code. This abstract presents two versions of an RTM simulation, reducing floating-point operations by changing the equation derivatives and improving the energy efficiency of the simulation. The paper studies optimizations in different multicore and GPU architectures, investigating the impact of other APIs on the energy efficiency and portability of the code. The energy efficiency results showed that the CUDA implementation running on the Volta architecture has the best energy efficiency for RTM on GPUs, while for multicore, the OpenACC version was the best.

Summary is not available.

Summary is not available.

Summary is not available.