Molecular dynamics (MD) simulations are computationally intensive, making energy efficiency a critical concern. Traditional optimization focuses on reducing either runtime or energy consumption, often overlooking their combined effect. This thesis introduces two new metrics—Energy Delay Product (EDP) and Energy per Floating Point Operation (EPF)—to assess energy efficiency in AutoPas, a performance-tuned MD simulation framework. By benchmarking various configurations, we examine how these metrics impact tuning and optimization strategies.

Our results show that optimizing for EDP achieves the best balance between execution time and energy consumption, leading to more efficient MD simulations. Notably, configurations with the shortest runtime also exhibit the lowest power consumption, aligning well with EDP. In contrast, EPF often leads to suboptimal setups by emphasizing computational throughput at the expense of higher energy usage. These findings suggest that conventional energy efficiency assessments in high-performance computing (HPC) may not accurately reflect real computational efficiency.

By integrating these new metrics into MD simulation evaluations, this study enhances energy-aware tuning in AutoPas and offers a more holistic approach to performance optimization. The results contribute to improving energy efficiency in MD simulations and HPC systems.