Search

• Adaption of Skylake architecture with AVX-512 support for SeisSol simulator. • Only utilized data obtained during SC18 Student Cluster Competition. • Good consistency is found between the results in the original paper and that in our reproduction. A SC'17 paper entitled Extreme Scale Multi-Physics Simulations of the Tsunamigenic 2004 Sumatra Megathrust Earthquake implemented a new cache-aware wave propagation scheme and an optimization approach for dynamic rupture using code generation. With the clustered local time-stepping method, they achieved a 13.6 × speedup for the Sumatra scenario of the simulation code SeisSol. In this work, we reproduce the optimization approach above on Intel Skylake, a relatively new architecture, to analyze its performance. We compare the performance results obtained on our platform with those in the original paper and study the differences with a profiling-based performance tool, Arm Forge. We perform scalability experiments on our cluster with the provided dataset. The speedup is consistent with the claims in the original paper. Also, a full 500-second simulation of the Sumatra earthquake on a small mesh is successfully reproduced on our platform. We visualize the results, and the process of how an earthquake develops is correctly revealed. Additionally, we analyze the impact of hyper-threading on the performance. [ABSTRACT FROM AUTHOR]