Evaluating the Energy Efficiency of Toubkal HPC: Insights for the Green500 Rankings and Its Metric

Energy efficiency has emerged as a vital metric for evaluating highperformance computing (HPC) systems, often measured by the Green500 rankings using the "performance per watt" criterion.In this paper, we assess the energy efficiency of the Toubkal Supercomputer based on the established Green500 metric.While we acknowledge the notable advancements in the rankings, we argue that the current energy efficiency metric for HPC systems adopted by the Green500 community overlooks two critical factors: the integration of on-site renewable energy sources and the potential for energy management through dynamic frequency scaling of processors.To support our argument, we consider GPU as well as dynamic voltage and frequency scaling (DVFS) technologies in a series of experiments.Utilizing our custom machine learning-based frequency predictor, our findings clearly illustrate the significant potential of energy management and DVFS to reduce energy consumption without compromising operational performance.