-
-
networking, network design, green software principles -
Alizadeh, N., Belchev, B., Saurabh, N. Kelbert, P., Castor, F. 2024. Analyzing the energy and accuracy of LLMs in software development. https://arxiv.org/pdf/2412.00329 -
https://arxiv.org/pdf/2412.00329 -
Examines tradeoffs between model accuracy and energy consumption for LLMs.
-
-
Alizadeh et al. 2025: Language Models in Software Development Tasks: An Experimental Analysis of Energy and Accuracy.
-
LLM, AI -
Alizadeh, N., Belchov, B., Saurabh, N., Kelbert, P., Castor, F. 2025. Language Models in Software Development Tasks: An Experimental Analysis of Energy and Accuracy. https://arxiv.org/abs/2412.00329 -
https://arxiv.org/pdf/2412.00329 -
Exploration of the trade-offs between model accuracy and energy consumption. The main findings are that higher energy coinsumption does not always lead to higher accuracy. Quantized versions of models can be both more efficient and more accurate compared to full-precision versions of medium sized models. No single model is suitable for all tasks in software development.
-
-
-
AI, genAI -
Bashir, N., Donti, P., Cuff, J., Sroka, S., Ilic, M., Sze, V., Delimitrou, C., Olivetti, E. 2023. The climate and sustainability implications of Generative AI. An MIT Exploration of Generative AI, March. https://doi.org/10.21428/e4baedd9.9070dfe7. -
https://mit-genai.pubpub.org/pub/8ulgrckc/release/2 -
An evaluation of our currently unsustainable approach toward Gen-AI’s development, underlining the importance of assessing technological advancement alongside the resulting social and environmental impacts.
-
-
Castano et al. 2023. Exploring the carbon footprint of Hugging Face's ML models: a repository mining study.
-
AI, Hugging Face, carbon -
Castano, J., Martinez-Fernandez, S., Franch, X., Bogner, J. 2023. Exploring the carbon footprint of Hugging Face's ML models: a repository mining study. https://arxiv.org/abs/2305.11164 -
https://arxiv.org/abs/2305.11164 -
The paper aims to analyze the measurement of the carbon footprint of 1,417 ML models and associated datasets on Hugging Face, which is the most popular repository for pretrained ML models
-
-
Chen et al. 2023 FrugalGPT: how to use large language models while reducing cost and improving performance
-
AI, frugal AI, GPT, green computing -
Chen, L., Zaharia, M., Zou, J. 2023. FrugalGPT: how to use large language models while reducing cost and improving performance. https://arxiv.org/pdf/2305.05176 -
https://arxiv.org/pdf/2305.05176 -
Presents and discusses three strategies that can lower the cost of LLM inference, specifically a) adapting prompts, b) LLM approximation, c) LLM cascade. The authors present an example called FrugalGPT that uses a simple LLM cascade strategy.
-
-
Domingo-Reguero et al. 2025: Energy-efficient neural network training through runtime layer freezing, model quantization, and early stopping
-
early stopping, AI, carbon, quantization -
Domingo-Reguero, A., Martinez-Fernandez, S., Verdecchia, R. 2025. Energy-efficient neural network training through runtime layer freezing, model quantization, and early stopping. Computer Standards and Interfaces, vol 92:103906. -
https://www.sciencedirect.com/science/article/pii/S0920548924000758?via%3Dihub -
Early stopping is the best tested method to save energy with minimal accuracy effect.
-
-
-
networking, network design, green software principles -
Nicolas Drouant, Éric Rondeau, Jean- Philippe Georges, and Francis Lepage. Designing green network architectures using the ten commandments for a mature ecosystem. Computer Communications, 42:38–46, 2014 -
https://hal.science/hal-00953000 -
Applies the "ten commandments" from ecology (specifically from Benyus, 2002: Biomimicry) to green network architecture design.
-
-
-
model, calculation, carbon footprint, tooling -
Faiz, A., Kaneda, S., Wang, R., Osi, R. 2024. Modeling the end-to-end carbon footprint of large language models. https://arxiv.org/html/2309.14393v2 -
https://arxiv.org/html/2309.14393v2 -
Describes a model for estimating LLM carbon emissions.
-
-
-
AI, image classification, carbon
-
Gonzalez Alvarez, A., Castano, J., Franch, X., Martinez-Fernandez, S. 2024. Impact of ML optimization tactics on greener pre-trained ML models. https://arxiv.org/pdf/2409.12878. -
https://arxiv.org/pdf/2409.12878 -
This study aims to analyze image classification datasets and pretrained models, improve inference efficiency by comparing optimized and non-optimized models, and assess the economic impact of the optimizations.
-
-
Hall, et al., 2021. Carbon-Aware Computing for Data Centers with Probabilistic Performance Guarantees.
-
carbon aware, data center
-
Hall, S., Micheli, F., Belgioso, G., Radovanovic, A., Dorfler, F. 2021. Carbon-Aware Computing for Data Centers with Probabilistic Performance Guarantees. https://arxiv.org/pdf/2410.21510 -
https://arxiv.org/pdf/2410.21510 -
The authors propose using day-ahead planning and real-time job placement to reduce energy consumption in dat centers.
-
-
Hoffman, et al., 2024. Improving Carbon Emissions of Federated Large Language Model Inference through Classification of Task-Specificity.
-
carbon footprint, machine learning, AI -
Hoffman, G-D., Majuntke, V. 2024. Improving Carbon Emissions of Federated Large Language Model Inference through Classification of Task-Specificity. HotCarbon’24, July 9, 2024, Santa Cruz, CA. https://hotcarbon.org/assets/2024/pdf/hotcarbon24-final109.pdf -
https://hotcarbon.org/assets/2024/pdf/hotcarbon24-final109.pdf -
The authors analyzed the differences between eight different LLMs. They find that they can classify and route prompts to the most energy efficient LLM in a federation of LLMs.
-
-
-
AI -
Huang, J., Gopal, S. 2025. Green AI - a multidisciplinary approach to sustainability. Environmental Science and Ecotechnology, Vol 24, 100536. -
https://www.sciencedirect.com/science/article/pii/S2666498425000146?via%3Dihub -
Correspondence article proposing framework for Green AI.
-
-
Jagannadharao et al. 2024. A beginner's guide to power and energy measurement and estimation for computing and machine learning.
-
power, measurement, models -
Jagannadharao, A., Beckage, N., Biswas, S., Egan, H., Gafur, J., Metsch, T., Nafus, D., Raffa, G., Tripp, C. 2024. A beginner's guide to power and energy measurement and estimation for computing and machine learning. https://arxiv.org/pdf/2412.17830. -
https://arxiv.org/pdf/2412.17830 -
Detailed overview of power measurement and estimation for computing, including machine learning.
-
-
Lee, H-P, et al. 2025. The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers.
-
Gen AI, human impacts -
Lee, H-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., Wilson, N. 2025. The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. https://www.microsoft.com/en-us/research/uploads/prod/2025/01/lee_2025_ai_critical_thinking_survey.pdf?ref=404media.co -
Evidence that use of GenAI diminishes critical thinking in knowledge workers.
-
-
-
AI, water -
Li, P., Yang, J., Islam, M., Ren, S., 2023. Making AI less thirsty - uncovering the secret water footprint of AI. https://arxiv.org/abs/2304.03271 -
Position paper explaining the need to include water in environmental imapct assessments for AI.
-
-
-
BLOOM, LLM, AI, calculation, model, carbon footprint -
Luccioni, A.S., Viguier, S., Ligozat, A-L. 2022. Estimating the Carbon Footprint of BLOOM, a 176B Parameter Language Model., https://arxiv.org/abs/2211.02001 -
Quantifies the carbon footprint of the BLOOM model across its life cycle, with the upper estimate being ~50.5 T.
-
-
-
AI, power -
Luccioni, A.S., Jernite, Y., Strubell, E. 2024. Power Hungry Processing: Watts Driving the Cost of AI Deployment?., https://arxiv.org/pdf/2311.16863 -
The authors measure the amount of energy and carbon required to perform 1000 inferences using several models. They find that multi-purpose generative architectures are far mor expensive than task specific versions. They caution that model utility should be "more intentionally weighed against increased costs in terms of energy and emissions".
-
-
-
carbon footprint, machine learning, AI -
Patterson, D., Gonzalez, J., Hölzle, U., Le, Q., Liang, C., Munguia, L-M., Rothchild, D., So, D., Texier, M., Dean, J. 2022. The carbon footprint of machine learning training will plateau then shrink, https://arxiv.org/pdf/2204.05149 -
https://arxiv.org/pdf/2204.05149 -
Describes four best poractises that can reduce energy used to train machine learning models by up to 100x and carbon emissions by 1000x.
-
-
-
carbon footprint, green computing -
Pazienza, A., Baselli, G., Carlo, D.C., Trussoni, M.V. 2024. A holistic approach to environmentally sustainable computing. Innovations in Systems and Software Engineering, 20: 347-371, https://link.springer.com/article/10.1007/s11334-023-00548-9 -
https://link.springer.com/article/10.1007/s11334-023-00548-9 -
Proposes the Environmentally Sustainable Computing framework and describes use-cases.
-
-
-
carbon awareness, data center -
Koningstein R, Schneider I, Chen B, Duarte A, Roy B, et al. Carbon-aware computing for datacenters. IEEE Trans Power Syst 2023;38(2):1270–80. -
http://dx.doi.org/10.1109/TPWRS.2022.3173250, https://ieeexplore.ieee.org/document/9770383. -
Describes Google's system for carbon-intelligent compute management. This is a system for scheduling workloads to minimize carbon footprints.
-
-
Ren et al. 2024: Reconciling the contrasting narratives on the environmental impact of large language models
-
carbon awareness, data center -
Ren et al 2024 Reconciling the contrasting narratives on the environmental impact of large language models, Sci Rep 14, 26310 (2024). https://doi.org/10.1038/s41598-024-76682-6 -
https://www.nature.com/articles/s41598-024-76682-6?error=cookies_not_supported&code=1ef76dbf-5291-4fe6-8dc6-57e92d6e9550#article-info -
Compares human and AI carbon emissions and discusses contrasting narratives.
-
-
-
carbon awareness, load shifting, scheduling, data centers -
Riepen, I., Brown, T., Zavala, V.M. 2024. Spatio-temporal load shifting for truly clean computing, Advances in Applied Energy, vol 17: 100202 -
https://www.sciencedirect.com/science/article/pii/S2666792424000404?via%3Dihub#sec1 -
Load shifting between regions and times of day can be effective at reducing carbon emissions for compute tasks. The optimum strategy for time and space shifting varies between regions and times of year. Carbon efficiency also reduces cost - applying optimal load shifting strategies reduced compute cost by ~ 1.3 EUR/MWh.
-
-
Roque et al. 2024. Unveiling the Energy Vampires: A Methodology for Debugging Software Energy Consumption
-
calculation, green computing -
Roque, E.B., Cruz, L., Durieux, T. 2024. Unveiling the Energy Vampires: A Methodology for Debugging Software Energy Consumption. https://arxiv.org/pdf/2412.10063 -
https://arxiv.org/pdf/2412.10063 -
Presents an energy debugging methodology for identifying and isolating energy consumption hotspots in software systems.
-
-
Tasdelen et al. 2024. Enhancing green computing through energy-aware training: an early stopping perspective
-
green computing -
Tasdelen, A., Enhancing green computing through energy-aware training: an early stopping perspective., Current Trends in Computing, Vol 2, 2: 108-139. -
https://dergipark.org.tr/en/download/article-file/4407207 -
This study examines energy efficient model training strategies and particularly highlights the role of early-stopping.
-
-
Tkachenko, 2024: Integrating AI’s Carbon Footprint into Risk Management Frameworks: Strategies and Tools for Sustainable Compliance in Banking Sector
-
accounting, reporting, regulation, AI, carboin footprint -
Tkachenko, 2024: Integrating AI’s Carbon Footprint into Risk Management Frameworks: Strategies and Tools for Sustainable Compliance in Banking Sector, https://arxiv.org/pdf/2410.01818 -
https://arxiv.org/pdf/2410.01818 -
Examines the integration of AI carbon emissions into risk management frameworks in banking. The paper describes how banks can identify, assess, and mitigate the carbon emissions associated with AI within their riskmanagement frameworks, including choosing energy-efficient models, using green cloud computing, and implementing lifecycle management. Advocates aligning with global standards and points out how this can ease regulatory compliance.
-
-
-
AI, carbon footprint -
Varoquaux, G., Luccioni, A.S., Whittaker, M. 2024. Hype, Sustainability, and the Price of the Bigger-is-Better Paradigm in AI. https://arxiv.org/pdf/2409.14160 -
https://arxiv.org/pdf/2409.14160 -
Scrutinizes the trends and trade-offs in scaling AI and refutes two common assumptions underlying the ‘bigger-is-better’ AI paradigm: 1) performance improvements result from increased scale, and 2) large-scale models are required to solve all interesting problems. The paper argues that approach is "fragile scientifically" and has negative externalities.
-
Weisner et al 2021 Let’s Wait Awhile: How Temporal Workload Shifting Can Reduce Carbon Emissions in the Cloud https://arxiv.org/pdf/2110.13234Carbon-Aware Computing for Data Centers with Probabilistic Performance Guaranteesa
-
Weisner, P., et al. 2021. Let's wait awhile: how temporal workload shifting can reduce carbon emissions in the cloud.
-
scheduling, cloud, carbon-aware -
Weisner, P., Behnke, I., Scheinert, D., Gontarska, K., Thamsen, L. 2021. Let's wait awhile: how temporal workload shifting can reduce carbon emissions in the cloud., https://arxiv.org/pdf/2110.13234 -
https://arxiv.org/pdf/2110.13234 -
The authors examine the potential impact of shifting computational workloads towards times where the energy supply is expected to be less carbon-intensive.
-
-
Wolff Anthony et al. 2007. CarbonTracker: tracking and predicting the carbon footprint of training deep learning models.
-
calculation, accounting, -
Wolff Anthony, L.E., Kanding, B., Selvan, R. 2007. CarbonTracker: tracking and predicting the carbon footprint of training deep learning models. https://arxiv.org/pdf/2007.03051 -
https://arxiv.org/pdf/2007.03051 -
A tool for tracking and predicting the energy and carbon footprint of training DL models
-
-
Wu et al. 2025. Unveiling Environmental Impacts of Large Language Model Serving: A Functional Unit View
-
ai, LLMs, functional unit -
Wu, Y., Hua, I., Ding, Y. 2025. Unveiling Environmental Impacts of Large Language Model Serving: A Functional Unit View. https://arxiv.org/pdf/2502.11256 -
https://arxiv.org/pdf/2502.11256 -
The authors propose a framework called FUEL, a functional unit-based way to assess the environmental impact of serving LLMs.
-
-
Zheng et al. 2020. Mitigating curtailment and carbon emissions through load migration between data centers
-
curtailment, load shifting, carbon awareness, data centers -
Zheng J, Chien AA, Suh S. Mitigating curtailment and carbon emissions through load migration between data centers. Joule 2020;4(10):2208–22. -
http://dx.doi.org/10.1016/j.joule.2020.08.001 https://www.sciencedirect.com/science/article/pii/S2542435120303470. -
Load migration can reduce renewable curtailment and GHG emissions. Existing data centers in the CAISO region can reduce up to 239 KtCO2e per year. Net abatement cost can largely stay negative
-
Links that still need sorting and formatting:
...