Short-Interval Energy Measurements for Algorithm Selection in Particle Simulations

Energy efficiency in High-Performance Computing (HPC) has become increasingly important in recent years. Particle simulations offer a wide range of algorithmic choices, and the optimal algorithmic configuration (AC) may differ depending on whether runtime or energy consumption is minimized. Due to Dynamic Voltage and Frequency Scaling (DVFS), modern CPUs adjust their operating frequency and voltage during less compute-intensive phases, leading to a non-linear relationship between runtime and energy consumption. In this work, we demonstrate that algorithm selection in particle simulations based on short-interval energy measurements from the Running Average Power Limit (RAPL) interface can be used to achieve energy savings.     «

Energy efficiency in High-Performance Computing (HPC) has become increasingly important in recent years. Particle simulations offer a wide range of algorithmic choices, and the optimal algorithmic configuration (AC) may differ depending on whether runtime or energy consumption is minimized. Due to Dynamic Voltage and Frequency Scaling (DVFS), modern CPUs adjust their operating frequency and voltage during less compute-intensive phases, leading to a non-linear relationship between runtime and ene...     »