Exploration of Energy Consumption Using the Intel Running Average Power Limit Interface

With the rising need for computational power, High Performance Computing (HPC) systems are finding their way into various frameworks such as Spaceborne applications. Achieving an optimum combination between performance and power utilization for HPCtype systems in harsh environments can be a challenge. In order to combat these challenges, it is essential to understand the characteristics of the system hardware and software. In this project, the Intel Running Average Power Limit (RAPL) interface provides measurements of energy consumption for two HPC applications, Weather Research and Forecasting (WRF) and TensorFlow. Applications were based on popularity in the HPC community and potential uses in space missions. A way to understand performance on a multi-core system is through affinity. Affinity, or pinning tasks and threads to cores, is needed to ensure workloads are distributed to the desired cores on the chip. Various affinity settings were applied to the WRF and TensorFlow applications in order to assess the performance and power efficiency at the single node level.