Energy efficiency — the number of instructions a computer can execute using one joule of energy — is improving every day, but at a slower rate than predicted by Koomey’s Law, which estimated that it would double every 19 months, rather than every 2.29 years as has been shown
UGR researchers Alberto Prieto and Beatriz Prieto, from the Department of Computer Engineering, Automation and Robotics (ICAR), together with Juan José Escobar, from the Department of Software Engineering, have published a study concluding that the energy efficiency of supercomputers doubles every 2.29 years. The study, in which Thomas Lampert of the University of Strasbourg also collaborated, has been published in the journal Cluster Computing (Springer Nature).
Energy efficiency is the number of instructions a computer can execute using one joule of energy. The authors show improvements in efficiency are being achieved at an exponential rate, thanks to advances in technology and air-conditioning systems in large data centres. However, these improvements are occurring at a slower rate than forecast by Koomey’s Law of 2009, which predicted that maximum energy efficiency would double at a faster rate, namely every 19 months, rather than the 27 months observed by the authors.
One of the main challenges facing society today is the need to reduce the demand for electricity. Information and communication technology (ICT) is an increasingly important driver of this type of energy consumption and exerts a significant impact on greenhouse gas emissions. Indeed, in 2015, the US Semiconductor Industry Association noted that while global energy production is growing linearly, the demand for electricity from computers is growing exponentially — a clear mismatch. Other studies from the same year suggested that, in the worst-case scenario, ICT could be responsible for up to 23% of global greenhouse gas emissions by 2030. If this trend were to continue in line with 2013 estimates, by 2040 the electrical energy consumed by ICT equipment would exceed the world’s total electrical energy production, meaning that there would not be enough energy to power all the world’s computer systems.
For ICT energy consumption to be sustainable, the energy efficiency of computing systems must grow at least as fast as the demand for computing services, including artificial intelligence applications. It is therefore crucial to understand how energy efficiency is evolving, and how it will evolve in the future, in order to take appropriate action. This is why the paper by these UGR researchers is particularly relevant.
The authors also demonstrate that computing performance (the number of instructions executed per second) continues to grow exponentially, doubling every 1.85 years. This is in line with Moore’s Law, which in 1975 established that this doubling would occur approximately every 2 years. Currently, the world’s most powerful supercomputer (Frontier) is capable of executing 1.2 quintillion operations with real numbers per second.
To reach these conclusions, the authors analysed a total of 9,682 high-performance computers (HPC) included in thirty TOP500 lists published between 2008 and 2023. The TOP500 ranks the 500 most powerful supercomputers in the world at any time and is published twice a year, in June and September.
This study was carried out in the context of the HPEEC-BIOBRAIN project, grant PID2022-137461, financed by MICIU/AEI/10.13039/501100011033 and ERDF/EU.
Full publication: Prieto, A., Prieto, B., Escobar, J. J., & Lampert, T. (2025). Evolution of computing energy efficiency: Koomey’s law revisited. Cluster Computing, 28(1), 42.
Contact details:
Alberto Prieto Espinosa (@email)
Translated version: This text has been translated into English by the Language Services Unit (Vice-Rectorate for Internationalization) of the University of Granada.
