Quantifying the Energy Cost of Performance Inefficiency in HPC Applications

Abstract

HPC systems consume a significant amount of energy, and inefficiencies in application performance further exacerbate this demand, increasing operational costs and environmental impact. As data center energy usage continues to rise with growing computational workloads, improving the energy efficiency of HPC applications has become a critical priority. This work introduces the Energy-Efficiency Critical Path Tool (EE-CPT), a profiling framework that quantifies the energy costs associated with load balancing and communication-related inefficiencies. Using ICON-A, a production-grade climate modeling application, as a case study, we demonstrate how EE-CPT correlates the traditional time-based performance metrics with energy consumption across different process, thread, and CPU frequency configurations. The results reveal optimal configurations that can balance runtime and energy use while identifying performance bottlenecks that offer opportunities for energy-aware optimization.

Authors 2

1. Radita Liem Aachen IT centre

   RWTH Aachen University

   Affiliation as printed

   IT Center, RWTH Aachen University, Aachen, Germany

2. Dlyaver Djebarov Aachen IT centre

   RWTH Aachen University

   Affiliation as printed

   IT Center, RWTH Aachen University, Aachen, Germany