Sustainable AI: Environmental Implications, Challenges and Opportunities

By Carole-Jean Wu, Ramya Raghavendra, Udit Gupta, Bilge Acun, Newsha Ardalani, Kiwan Maeng, Gloria Chang, Fiona Aga Behram, J. Huang, Charles Bai and 15 more. 2021

662 citations recorded by the source corpus.

This paper explores the environmental impact of the super-linear growth trends for AI from a holistic perspective, spanning Data, Algorithms, and System Hardware. We characterize the carbon footprint of AI computing by examining the model development cycle across industry-scale machine learning use cases and, at the same time, considering the life cycle of system hardware. Taking a step further, we capture the operational and manufacturing carbon footprint of AI computing and present an end-to-end analysis for what and how hardware-software design and at-scale optimization can help reduce the overall carbon footprint of AI. Based on the industry experience and lessons learned, we share the key challenges and chart out important development directions across the many dimensions of AI. We hope the key messages and insights presented in this paper can inspire the community to advance the field of AI in an environmentally-responsible manner.

Subfields: Computer Networks and Communications, Computer Science Applications, Electrical and Electronic Engineering, AI Ethics, Fairness & Society

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Networks and Communications

In Computer Science Applications

In Electrical and Electronic Engineering

In AI Ethics, Fairness & Society

Citations in the corpus

LLaMA: Open and Efficient Foundation Language Models (2023, methodology): “We follow a formula for Wu et al. (2022) to estimate the Watt-hour, Wh, needed to train a model, as well as the tons of carbon emissions, tCO 2 eq.”
Llama 2: Open Foundation and Fine-Tuned Chat Models (2023, methodology): “Following preceding research (Bender et al., 2021a; Patterson et al., 2021; Wu et al., 2022; Dodge et al., 2022) and using power consumption estimates of GPU devices and carbon efficiency, we aim to calculate the carbon emissions resulting from the pretraining of Llama 2models.”
Efficient Large Scale Language Modeling with Mixtures of Experts (2021, neutral): “…for our dense and MoE models, respectively, based on observed training speeds for our largest models.14 We note that these estimates do not account for the costs associated with manufacturing the infrastructure to train such models, which can be significant (Gupta et al., 2021; Wu et al., 2021).”
Evaluating the Social Impact of Generative AI Systems in Systems and Society (2023, background): “…There is still a lot of uncertainty around certain variables, such as the relative contribution of added parameters to their energy consumption and carbon footprint, as well as the proportion of energy used for pre-training versus fine-tuning models for different tasks and architectures [267].”
The Flan Collection: Designing Data and Methods for Effective Instruction Tuning (2023, background): “On the contrary, pretrained models that require extensive finetuning become more costly when aggregating over many millions of additional training steps (Wu et al., 2022; Bommasani et al., 2021).”
Measuring the Carbon Intensity of AI in Cloud Instances (2022, background): “Other studies have gone beyond simply tracking the emissions from training models, aiming to quantify the emissions resulting from manufacturing computing hardware [15], the broader impacts of sustainable AI [49], and the methodologies used to assess those impacts [21, 26].”
Exploring AI Ethics of ChatGPT: A Diagnostic Analysis (2023, background): “In addition, the increasing demand of computational resources is leading to the energy consumption and carbon emission, negatively impacting the environment [97].”
A Holistic Approach for Designing Carbon Aware Datacenters (2022, background): “Latency-tolerant workloads, such as batch and AI training jobs [57], [67], target specific SLO categories that include 4, 8- and 24-hour completion times.”
Estimating the Carbon Footprint of BLOOM, a 176B Parameter Language Model (2022, background): “Other studies have pursued a broader analysis of trends in terms of the energy requirements and CO2 emissions of ML models in general (Thompson et al., 2020; Wu et al., 2021; Patterson et al., 2022).”
Power Hungry Processing: Watts Driving the Cost of AI Deployment? (2023, neutral): “…approximately one-third of their internal end-to-end ML carbon footprint to model inference, with the remainder produced by data management, storage, and training [57]; similarly, a 2022 study from Google attributed 60% of its ML energy use to inference, compared to 40% for training [40].”
Aligning artificial intelligence with climate change mitigation (2022, background): “As data centres become increasingly efficient and powered by clean electricity, the relative importance of emissions from non-operational life-cycle phases will grow— particularly embodied emissions in computing hardware and data centre building construction 33,74 .”
ACT: designing sustainable computer systems with an architectural carbon modeling tool (2022, background): “Within the domain of AI, researchers have quantified the impact of large neural network training on carbon footprints [37, 59, 96, 103].”
FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance (2023, neutral): “In addition to the ﬁnancial burden, employing larger LLMs incurs signiﬁcant environmental and energy impacts Bender et al. (2021); Wu et al. (2022), affecting the social welfare of current and future generations.”
RecShard: statistical feature-based memory optimization for industry-scale neural recommendation (2022, background): “A significant fraction of deep learning compute cycles in industry-scale data centers can be attributed to deep learning recommendation models (DL-RMs) [3, 4, 9, 11, 20, 34, 43, 47, 48].”
A systematic review of Green AI (2023, background): “[106] provide a landscape of the carbon footprint of AI models across Facebook.”