This Is Taking Too Long - Investigating Time as a Proxy for Energy Consumption of LLMs
Lars Krupp, Daniel Geißler, Francisco M. Calatrava-Nicolas, Vishal Banwari, Paul Lukowicz, Jakob Karolus · 2026
Read This Is Taking Too Long - Investigating Time as a Proxy for Energy Consumption of LLMs by Lars Krupp, Daniel Geißler, Francisco M. Calatrava-Nicolas, Vishal Banwari, Paul Lukowicz, Jakob Karolus (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-15699.txt.
Description
Krupp and co-authors probe whether run time can stand in for energy use. API language models hide their hardware and their power draw. Users see only slow or fast replies. The team compares time-based estimates with real meter readings. Locally hosted twin models supply the ground truth. Timing patterns even reveal which GPU an API model runs on. That clue grounds firmer energy estimates for end users. The work targets a clearer view of hidden inference costs.
Catalog record
| CREATOR | Lars Krupp, Daniel Geißler, Francisco M. Calatrava-Nicolas, Vishal Banwari, Paul Lukowicz, Jakob Karolus |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 26312 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.15699