Shelby Hall Graduate Research Forum Posters

Files

Download Available for download on Thursday, June 03, 2027

Download Full Text (508 KB)

Description

This study systematically evaluates 14 carbon and energy tracking technologies to explore the features and limitations of each tool and evaluate their effectiveness in precisely quantifying power consumption and carbon emissions using a series of standardized AI benchmarking algorithms. Two machine learning models are designed to simulate growing computing loads, a Deep Neural Network (DNN) with five fully linked layers and a Convolutional Neural Network (CNN) with several intricate layers and run on local systems with a uniform power usage effectiveness (PUE) of 1.5 and a carbon intensity of 393 CO2/ kWh unless otherwise stated. Metrics such as overall power consumption (kWh), projected carbon emissions, and utilization of individual devices (CPU, GPU, RAM) are standardized for direct comparison. The interoperability, usefulness in Python environments, and consistency across code versions revealed differences in accuracy and sensitivity of each tool are analyzed in this paper. This study offers practical guidance on the selection of carbon tracking technologies for AI applications, assisting engineers in making data-driven decisions to improve sustainable AI development processes.

Publication Date

3-2025

Department

Systems Engineering

City

Mobile

Disciplines

Artificial Intelligence and Robotics | Computer and Systems Architecture | Industrial Technology | OS and Networks | Other Computer Sciences | Other Operations Research, Systems Engineering and Industrial Engineering | Systems Engineering

Comparative Analysis of Tools to Track Energy and Carbon Emissions for AI Computation

Share

COinS