Long-Term Analysis of Summit’s Power & Thermal Dynamics (2020 – 2021)

Overview

This project analyzed extensive operational data to understand Summit’s power, energy, and thermal behavior, leveraging insights from the Cooling Intelligence for Summit initiative. The findings provided a foundation for ML-based methods in energy efficiency.

• Conducted a comprehensive study on power consumption at component, node, and system levels across all 4,626 Summit compute nodes.
• Analyzed over 840,000 Summit jobs and 250,000 GPU failure logs, uncovering operational insights.
• Processed high-frequency 1Hz telemetry data spanning the entire year of 2020.
• Authored and led a best paper award-winning study at SC21, the premier conference in HPC.
• Research team received the UT-Battelle Research Accomplishment Award (2022) for contributions to HPC energy efficiency.

Technology

• OLCF Summit Supercomputer Telemetry (2021)
• OLCF Andes cluster
• Dask, parquet, pandas
• Jupyter, seaborn

Publication

• Shin et al., “Revealing power, energy and thermal dynamics of a 200PF pre-exascale supercomputer”, SC21, 2021