Exploring the energy-time tradeoff in high-performance computing

High-performance computing is and has always been performance oriented. However, a consequence of the push towards maximum performance is increased energy consumption, especially at supercomputing centers. Moreover, as peak performance is rarely attained, some of this energy consumption results in little or no performance gain. In addition, large energy consumption costs the government a significant amount of money and wastes natural resources. This paper investigates the tradeoff between energy and performance. Through the use of processors that support frequency and voltage scaling, we measured the performance and energy consumption of programs from three popular benchmark sets. We took multiple measurements for each program using different frequency and voltage settings. Results show that for many programs, a significant decrease in energy is possible with a small increase in time. We believe that this justifies further investigation into parallel HPC (e.g., MPI) applications.