More than Carbon: Cradle-to-Grave environmental impacts of GenAI training on the Nvidia A100 GPU

TL;DR

Results for GPT-4 training show that the use phase dominates 10 categories, contributing 96% to climate change and fossil fuel depletion, and the GPU chip is the largest contributor in 10 categories, particularly climate change and fossil resource use.

Abstract

The rapid expansion of Artificial Intelligence (AI) has intensified concerns about its environmental sustainability. Current assessments focus on operational carbon emissions using secondary data, overlooking impacts in other life cycle stages. This study presents a comprehensive multi-criteria life cycle assessment of AI training, examining 16 environmental impact categories using primary data from the Nvidia A100 SXM 40 GB GPU. Results for GPT-4 training show that the use phase dominates 10 categories, contributing 96% to climate change and fossil fuel depletion. Manufacturing dominates 6 categories, including human toxicity (94%) and freshwater eutrophication (81%). The GPU chip is the largest contributor in 10 categories, particularly climate change (81%) and fossil resource use (80%). While primary data produces modest changes in carbon estimates, substantial variations emerge elsewhere, e.g. minerals and metals depletion increases by 33%. This analysis expands the Sustainable AI discourse beyond carbon emissions, challenging current sustainability narratives.

Figures (9)

• Figure 1: Modeling principles of this study. (a) Generic technology context for the product system and foreground system. (b) Data collection methodology. (c) Background system. (d) Life cycle stages. (e) Allocation for the reference flow. (f) Functional unit. (g) Two different configurations considered in this study to illustrate the model on real-life case studies (author illustration).
• Figure 2: Annotated components of the Nvidia SXM A100 40 GB GPU showing which components were the subject of a detailed teardown analysis and which underwent an elemental analysis via ICP-OES (author pictures).
• Figure 3: Product environmental impact assessment of one GPU card. Impact contribution by life cycle stage across 16 environmental impact categories. Numbers within bars indicate the total absolute environmental impact values as sum of all life cycle stages (author illustration).
• Figure 4: Environmental impact distribution across GPU component categories (casing, heatsink, PCB, GPU chip (main die + VRAM), PoP, and upstream distribution) for the cradle-to-gate stages of the Nvidia A100 GPU. Numbers within bars indicate the total absolute environmental impact values as sum of all component categories (author illustration).
• Figure 5: Impact contribution by life cycle stage for 16 environmental impact categories by AI model training BLOOM and GPT-4. Numbers within bars indicate the total absolute environmental impact values as sum of all life cycle stages (author illustration).