Architectural Adaptation and Performance-Energy Optimization for CFD Application on AMD EPYC Rome

The advantages of the second-generation AMD EPYC Rome processors can be successfully used in race to Exascale. However, novel architecture's complexity makes it challenging adapt demanding scientific codes - like stencil ones platforms with CPUs. This article tackles this challenge by exploring adaptation stencil-based CFD (computational fluid dynamics) application called MPDATA these processors' influential features. We show that previously proposed parametric methodology profitably applied extend performance portability memory-bound on architecture. extension architecture requires careful consideration two relevant aspects reflect splitting into multiple dies features cache hierarchy and partitioning cores work teams. also investigates correlation between optimizations energy efficiency for a ccNUMA platform powered top-of-the-line 64-core 7742 CPUs, comparing results against servers Intel Xeon Scalable different generations. Even without appealing prices, achieved are solid argument confirming competitiveness CPUs applications.