High-order ADER-DG minimizes energy- and time-to-solution of SeisSol

Abstract

In this paper we give a comprehensive overview of our nodelevel optimization of the high-order finite element software SeisSol aiming at minimizing energy- and time-to-solution. SeisSol simulates dynamic rupture and seismic wave propagation at petascale performance in production runs. In this context we analyze the impact that convergence order, CPU clock frequency, vector instruction sets and chip-level parallelism have on the execution time, energy consumption and accuracy of the obtained solution. From a performance perspective, especially on state-of-the-art and future architectures, the shift from a memory- to a compute-bound scheme and the need for double precision arithmetic with increasing orders of convergence is compelling. Our results show that we are able to reduce the computational error by up to five orders of magnitudes when increasing the order of the scheme from 2 to 7, while consuming the same amount of energy.