Home // International Journal On Advances in Systems and Measurements, volume 17, numbers 1 and 2, 2024 // View article

Capability and Applicability of Measuring AI Model’s Environmental Impact

Authors:
Rui Zhou
Tao Zheng
Xin Wang
Lan Wang
Emilie Sirvent-Hien
Nathalie Charbonniaud

Keywords: AI Environmental Impact; CO2e; FLoating point of OPerations (FLOPs); Power Usage Effectiveness (PUE); Pragmatic Scaling Factor (PSF); Thermal Design Power (TDP); Multiple Object Tracking Accuracy (MOTA).

Abstract:
More and more of Artificial Intelligence (AI) systems have been adopted by Information and Communication Technology (ICT) solutions to make effective digital transformation. In recent years environmental impact of AI systems has been investigated and methodologies have been developed to calculate their cost. In this paper, we survey, analyze, and evaluate three types of tools for counting the energy consumption/CO2 emission (CO2e) of AI systems. By verifying them in sets of experiments, including centralized and distributed on devices architecture, we compare ease of use of tools, simulation result vs real measurement and finally bring advice to help AI developers to take into account environmental cost of AI models with measurement tools. Finally, we developed a measurement tool for AI model environment impact based-on our experiments on the power consumption of AI models and applied our tool on AI model to verify optimization results.

Pages: 25 to 35

Copyright: Copyright (c) to authors, 2024. Used with permission.

Publication date: June 30, 2024

Published in: journal

ISSN: 1942-261x