News
Investigating the Impact of Measurement Frequency on Energy Readings
Our collaborative research paper on the analysis of measurement frequencies in energy readings for CPUs and GPUs has been accepted at the HotCarbon'2026 Workshop on Sustainable Computer Systems.
07/07/2026
We are excited to announce the acceptance of our collaborative paper, "Evaluating the Influence of Measurement Frequency on Energy Readings Using Intel RAPL and NVIDIA NVML," at the HotCarbon'2026 Workshop on Sustainable Computer Systems. This workshop focuses on sustainable computing systems, a field that increasingly relies on precise energy consumption measurements.
Working hand-in-hand with our colleagues from the Faculty of Computer Science and Mathematics , we investigated how various measurement intervals impact energy readings on CPUs and GPUs. Our research provides crucial insights into optimal measurement methods for effective energy management.
Key Takeaways:
- For NVIDIA NVML: Cumulative-energy counters can show frequent stale reads below 100 ms and display strong platform-dependent behavior. It is recommended to avoid unvalidated counters and prefer power integration.
- For Intel RAPL: Energy readings remain stable at sampling intervals of 10 ms or coarser, while sub-millisecond sampling mostly just increases overhead without a gain in accuracy.
A huge thank you to Prof. Dr. Andreas Brunnert, Manuel Steinberg, Maximilian Dauner, Dr. Benedikt Zönnchen, and Benedikt Schicker for their dedicated teamwork and collaboration.