How Green Can AI Be? A Study of Trends in Machine Learning Environmental Impacts

TL;DR

This study addresses whether AI can be truly green by quantifying environmental impacts across hardware production and model training over 2013–2023. It compiles a detailed dataset of NVIDIA workstation GPUs and couples hardware characteristics to training workloads via the Epoch AI dataset, computing life-cycle indicators with the MLCA framework. The findings show that production impacts and training footprints rise exponentially, and that rebound effects and impact shifting undermine the benefits of current decarbonization strategies, even with compute-location shifting. The work highlights the need for holistic lifecycle thinking and potentially reducing AI activity alongside efficiency improvements to meaningfully reduce AI’s environmental burden.

Abstract

The compute requirements associated with training Artificial Intelligence (AI) models have increased exponentially over time. Optimisation strategies aim to reduce the energy consumption and environmental impacts associated with AI, possibly shifting impacts from the use phase to the manufacturing phase in the life-cycle of hardware. This paper investigates the evolution of individual graphics cards production impacts and of the environmental impacts associated with training Machine Learning (ML) models over time. We collect information on graphics cards used to train ML models and released between 2013 and 2023. We assess the environmental impacts associated with the production of each card to visualize the trends on the same period. Then, using information on notable AI systems from the Epoch AI dataset we assess the environmental impacts associated with training each system. The environmental impacts of graphics cards production have increased continuously. The energy consumption and environmental impacts associated with training models have increased exponentially, even when considering reduction strategies such as location shifting to places with less carbon intensive electricity mixes. These results suggest that current impact reduction strategies cannot curb the growth in the environmental impacts of AI. This is consistent with rebound effect, where the efficiency increases fuel the creation of even larger models thereby cancelling the potential impact reduction. Furthermore, these results highlight the importance of considering the impacts of hardware over the entire life-cycle rather than the sole usage phase in order to avoid impact shifting. The environmental impact of AI cannot be reduced without reducing AI activities as well as increasing efficiency.

Figures (8)

• Figure 1: Evolution of the characteristics of NVIDIA workstation graphics cards from 2013 to 2023
• Figure 2: Evolution of the energy consumption of NVIDIA workstation graphics cards from 2013 to 2023
• Figure 3: Evolution of the production impacts (left in GWP, right in ADP) of NVIDIA workstation graphics cards.
• Figure 4: Evolution of the production impacts (left in GWP, right in ADP) of the graphics cards used for training systems in the Epoch AI data-set. Value intervals correspond to cases of ambiguous card names. The blue line represents the trend in the reference values while the red line represents the trend on the minimal values.
• Figure 5: Evolution of the number of graphics cards used for training models in the Epoch AI dataset. Trend was computed using the Weighted Least Square (WLS) method to account for heteroscedasticity