Pith. sign in

REVIEW 4 major objections 5 minor 63 references

Calculating Software's Energy Use and Carbon Emissions: A Survey of the State of Art, Challenges, and the Way Ahead

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A survey of 21 tools for calculating software energy and carbon emissions proposes a three-way taxonomy and identifies the gaps that prevent accurate reporting.

desk verdict A useful but rough survey of 21 energy/carbon tools; the taxonomy has a real inconsistency (CloudWatch is both Black-Box and Monitoring) that should be fixed before relying on it for tool selection. read the letter →

arxiv 2506.09683 v1 pith:XUMZ233X submitted 2025-06-11 cs.SE cs.CY

Priyavanshi Pathania , Nikhil Bamby , Rohit Mehra , Samarth Sikand , Vibhu Saujanya Sharma , Vikrant Kaulgud , Sanjay Podder , Adam P. Burden This is my paper
classification cs.SEcs.CY
keywords softwareenergymeasurementcarbonfootprintgreentaxonomyAIemissionstoolcomparisonpowermonitoringsustainability
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Software and AI training consume energy and emit carbon, but the tools meant to measure that impact are scattered and hard to compare. This survey examines 21 energy and carbon calculation tools and argues that every current approach falls into one of three categories: monitoring (direct measurement), estimation (model-based approximation), or black-box (proprietary cloud-provider tools). It builds a Tool Matrix that compares these tools by component coverage (CPU, GPU, RAM, storage, network), carbon intensity handling, and working mode, and it compiles the approaches component by component. On that basis the paper claims that the current state of the art leaves specific gaps—no process-level energy visibility in most tools, static rather than real-time grid intensity, neglected embodied emissions, heavy reliance on TDP for CPU, thin GPU estimation, and no standard SCI reporting. The intended contribution is a practical basis for tool selection and a research agenda for the green software community.

What carries the argument

The load-bearing objects are the proposed taxonomy and the Tool Matrix. The taxonomy is a three-branch classification of calculation mechanisms—Monitoring, Estimation, and Black-Box—defined by how a tool obtains energy and carbon numbers. The Tool Matrix is a table comparing 21 tools across attributes such as license, AI context, category, per-component approach (CPU, GPU, RAM, storage, network), carbon-estimation method, working mode, and supported OS. The component-wise consolidation (Table III) groups approaches by measured component. These structures carry the argument because they turn scattered tools into a comparable landscape, which is what supports the selection guidance and the challenge list.

What would settle it

Check the source code or documentation of any tool that the Tool Matrix lists as lacking process-level energy measurement and run it in process-tracking mode on a single process; if it correctly attributes energy to that process, the matrix entry would be wrong, and the survey's comparative claims would need revision. Alternatively, finding a published energy- or carbon-calculation tool for software that does not fit Monitoring, Estimation, or Black-Box categories would falsify the taxonomy's completeness.

Watch

Extended reading notes

Core claim

The paper's central claim is that the energy and carbon calculation landscape for software and AI can be organized by the mechanism a tool uses: Monitoring approaches measure energy directly through hardware power meters or software power meters such as RAPL-based counters; Estimation approaches approximate energy from coefficients, similarity to benchmarks, analytical power models, or custom-trained models; Black-Box approaches are proprietary cloud-provider calculators whose internals are undisclosed. The authors assert that this taxonomy, together with the Tool Matrix comparing 21 tools and the component-wise consolidation, reveals that no single tool covers all components and that several widely used shortcuts (TDP-based CPU estimation, static carbon intensity, ignoring embodied emissions) undermine accuracy. They further claim that the taxonomy is useful to practitioners selecting tools and to researchers targeting the identified challenges.

Load-bearing premise

The survey's conclusions rest on the assumption that the 21 tools it selected, drawn from academic papers, cloud-provider pages, and a green-software community repository, fairly represent the whole state of the art, and that the capabilities recorded in the Tool Matrix are accurate and current for each tool.

Editorial extensions

If this is right

  • A practitioner can use the taxonomy and Tool Matrix to choose a calculation approach that matches the required granularity and component coverage, rather than defaulting to one tool.
  • The component-wise consolidation shows that CPU and RAM are well covered by monitoring approaches, while storage and network rely mostly on coefficient estimates, so accuracy for those components is weakest.
  • The challenge list gives researchers a concrete agenda: process-level energy visibility, real-time grid intensity, embodied emissions, GPU-agnostic monitoring, and SCI reporting are the gaps to close.
  • Tool builders can consult the matrix to see what an ideal tool would need, including multi-experiment tracking and per-component coverage across CPU, GPU, RAM, storage, and network.
  • Because most tools do not report SCI, the survey implies that standardized software carbon-intensity reporting is still missing and is a necessary step for comparable environmental claims.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Inference (not in the paper): the taxonomy's three categories are general enough that likely any future software energy/carbon tool will fit one of them, so the taxonomy could become a common vocabulary for the field, not just a description of the 21 sampled tools.
  • Inference (not in the paper): if the identified gaps are real, then many published AI-training carbon numbers produced by the surveyed tools may be underestimates, since they typically exclude embodied emissions and use static grid intensity.
  • Inference (not in the paper): a natural testable extension of this survey is a reproducibility study that runs several documented workloads through each tool and checks whether measured outputs match the Tool Matrix entries, which would validate the matrix's accuracy.
  • Inference (not in the paper): the authors' stated plan to build an AI-lifecycle carbon calculation approach suggests that the challenge list doubles as a design specification; if their tool addresses process-level energy and real-time grid intensity, it could become a reference point for future tools.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. This manuscript presents a survey of 21 tools for calculating software energy use and carbon emissions. It proposes a three-way taxonomy of calculation approaches — Monitoring, Estimation, and Black-Box — describes representative tools in Sections II and III, compares them in a Tool Matrix (Table II) across attributes such as component coverage, carbon estimation, working mode, and operating system support, and provides a component-wise consolidation in Table III. Section V lists open challenges, including process-level energy visibility, real-time grid intensity, embodied emissions, GPU coverage, and SCI reporting. The paper is written primarily for practitioners and aims to support tool selection and future research in green software.

Significance. If the survey's coverage and taxonomy were reliable, this would be a useful practitioner-oriented reference: the Tool Matrix compiles 21 tools with license, approach, component coverage, carbon estimation, working mode, and OS support, and the challenge list in Section V identifies actionable gaps. A strength is that the individual tool descriptions and matrix entries are largely consistent with public documentation and can be independently verified. The main weaknesses are that the central taxonomy is applied inconsistently to the paper's own examples, the Black-Box category is not orthogonal to the other two categories, and the selection methodology is informal. These issues currently limit the paper's value as a definitive 'state-of-the-art' map, although they appear fixable within the manuscript's scope.

major comments (4)
  1. [§II.C and Table II] In §II.C, AWS CloudWatch Agent and Azure Machine Learning are introduced as examples of Black-Box approaches, yet Table II classifies Amazon CloudWatch as Monitoring (M, SB) using gpumon.py/nvidia-smi power-draw metrics, and Azure Machine Learning is labeled BB while its row reports the GpuEnergyJoules metric, which is a concrete monitoring capability. This inconsistency means that the taxonomy, as applied, cannot currently support the paper's claim that the Tool Matrix enables practitioners to select an appropriate calculation approach.
  2. [§II.C and Table I] The Black-Box category is defined by non-disclosure of internal methodology, which is a transparency property rather than a calculation mechanism; as such, it is not orthogonal to Monitoring and Estimation. This is visible in Table II, where CodeCarbon and eco2AI are marked as both M and E, showing that a single M/E/BB label does not partition the tool space. I recommend reworking the taxonomy into separate dimensions (e.g., measurement vs. estimation, and transparency/disclosure) and, if a single label is retained, applying it consistently to the examples.
  3. [§V, Dependence on approximate approaches] The statement that '50% of the TDP is considered to be the average power consumption' is asserted without a citation. Because this is a survey paper, an empirical claim about current practice needs a source or should be removed; as written, it is a load-bearing premise for the challenge that TDP-based estimation is inaccurate.
  4. [§II, Methodology] The selection procedure is described only as gathering tools from academic literature, Cloud Service Providers, the Green Software Foundation repository, and open exploration. For a survey claiming to cover the state of the art, this informal sampling does not establish that the 21-tool set is representative, and the Tool Matrix may silently omit relevant tools. Please document the search strategy, inclusion/exclusion criteria, screening process, and the date on which each tool entry was last verified against its public documentation.
minor comments (5)
  1. [§V, Limited use of real time values for RCI] The RCI variation example reports units as 'ktCO2eq/kWh'; this appears to be a typo, since values such as 41.5 and 73.9 in these units would be implausibly large for a regional grid intensity. The intended unit is likely gCO2eq/kWh.
  2. [§II.B.1 and §V] The paper says in §II.B.1 that TDP can be used as an approximation of component power consumption at maximum use, while §V states that TDP is not the maximum power the processor can consume and is not a good estimate of energy. These statements should be reconciled explicitly so that readers know whether TDP is being endorsed as a rough upper-bound proxy or rejected as an unreliable proxy.
  3. [Table II] Some cells in Table II, such as the entries for Green Algorithms and Cumulator, contain multiple similar values in adjacent columns (e.g., 'Static Grid Intensities' followed by a region-specific value); separating the energy and carbon columns more clearly would improve readability.
  4. [Table III] In the CPU Estimation row, 'Custom Model Training Based ESA VE' should be 'ESAVE' to match the cited tool name; the current spacing appears to be a typo.
  5. [§III, CarbonTracker] The description of CarbonTracker says it 'calculates and predicts' energy consumption and carbon footprint, but the Tool Matrix lists it as Monitoring; clarifying whether the predictive/forecasting component is considered part of the Monitoring approach or an Estimation component would avoid confusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the survey's taxonomy, Tool Matrix, and challenge list are descriptive compilations of externally documented tool features; the two self-citations are illustrative only and not load-bearing.

full rationale

The paper is a survey and makes no derived quantitative predictions, so there is no fitted-parameter or premise-conclusion chain that could reduce to its own inputs. Its central content is a taxonomy of existing energy/carbon calculation approaches (Monitoring, Estimation, Black-Box), a Tool Matrix of 21 tools, and a list of challenges. These are descriptive classifications drawn from external sources (RAPL, pyNVML, CodeCarbon, Cloud Jewels, CSP documentation, GSF materials, etc.) and are not derived from the paper's own assumptions. The two self-citations are used only as examples: [7] illustrates a potential 48% carbon saving from VM resizing in the introduction, and [33] is cited in Section II.B.4 as one instance of custom-model-training-based estimation. Neither supports the taxonomy, the matrix, or the challenge list, so no load-bearing argument reduces to a self-citation. The reviewer-noted inconsistency in applying the taxonomy (AWS CloudWatch appears both as a Black-Box example in Section II.C and as a Monitoring/SB tool in Table II) is an internal consistency or correctness concern about the taxonomy's application, not a circularity. Because the survey is self-contained against external tool documentation and no prediction is constructed from its own outputs, the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The paper introduces no fitted parameters or invented entities. It rests on three domain assumptions: the representativeness of the tool sample, the validity of component-wise decomposition, and the adequacy of its own taxonomy. These are listed above.

assumptions (3)
  • domain assumption The 21 surveyed tools are representative of the state of the art for software energy and carbon calculation.
    Section II describes informal gathering from academic literature, CSPs, GSF repository, and open exploration; no inclusion criteria are given. The survey's conclusions about gaps depend on this assumption.
  • domain assumption Energy consumption can be decomposed into CPU, GPU, RAM, Storage, and Network components, and component-wise comparison is a valid basis for tool selection.
    Section IV B and Table III use this decomposition without justifying its completeness for real systems.
  • ad hoc to paper The proposed Monitoring/Estimation/Black-Box taxonomy is adequate to classify existing calculation approaches.
    The taxonomy is introduced by the authors and not derived from a standard classification; some tools are assigned to multiple categories, so exclusivity is not established.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Calculating Software's Energy Use and Carbon Emissions: A Survey of the State of Art, Challenges, and the Way Ahead." pith.science (2026). https://pith.science/paper/XUMZ233X

@misc{pith2026250609683,
  author       = {Pith},
  title        = {Pith review of: Calculating Software's Energy Use and Carbon Emissions: A Survey of the State of Art, Challenges, and the Way Ahead},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XUMZ233X}},
  note         = {Machine review of arXiv:2506.09683}
}
read the original abstract

The proliferation of software and AI comes with a hidden risk: its growing energy and carbon footprint. As concerns regarding environmental sustainability come to the forefront, understanding and optimizing how software impacts the environment becomes paramount. In this paper, we present a state-of-the-art review of methods and tools that enable the measurement of software and AI-related energy and/or carbon emissions. We introduce a taxonomy to categorize the existing work as Monitoring, Estimation, or Black-Box approaches. We delve deeper into the tools and compare them across different dimensions and granularity - for example, whether their measurement encompasses energy and carbon emissions and the components considered (like CPU, GPU, RAM, etc.). We present our observations on the practical use (component wise consolidation of approaches) as well as the challenges that we have identified across the current state-of-the-art. As we start an initiative to address these challenges, we emphasize active collaboration across the community in this important field.

Figures

Figures reproduced from arXiv: 2506.09683 by the authors.

Figure 1
Figure 1. Proposed taxonomy of Energy and Carbon calculation approaches [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 59 canonical work pages

  1. [1]

    The real climate and transformative impact of ict: A critique of estimates, trends, and regulations,

    C. Freitag, M. Berners-Lee, K. Widdicks, B. Knowles, G. S. Blair, and A. Friday, “The real climate and transformative impact of ict: A critique of estimates, trends, and regulations,”Patterns, 2021

    work page 2021

  2. [2]

    Assessing ict global emissions footprint: Trends to 2040 & recommendations,

    L. Belkhir and A. Elmeligi, “Assessing ict global emissions footprint: Trends to 2040 & recommendations,”Journal of cleaner production, 2018

    work page 2018

  3. [3]

    Energy and policy consider- ations for modern deep learning research,

    E. Strubell, A. Ganesh, and A. McCallum, “Energy and policy consider- ations for modern deep learning research,” inProceedings of the AAAI conference on artificial intelligence, 2020

    work page 2020

  4. [4]

    Counting carbon: A survey of factors influencing the emissions of machine learning,

    A. S. Luccioni and A. Hernandez-Garcia, “Counting carbon: A survey of factors influencing the emissions of machine learning,”arXiv preprint arXiv:2302.08476, 2023

    arXiv 2023

  5. [5]

    Climate change 2022: Mitigation of climate change,

    IPCC, “Climate change 2022: Mitigation of climate change,” https://www.ipcc.ch/report/sixth-assessment-report-working-group-3/, Accessed - 11/07/2024

    work page 2022

  6. [6]

    Green software is a priority for software practitioners, but still lacks leadership support,

    S. of Green Software, “Green software is a priority for software practitioners, but still lacks leadership support,” https://stateof.greensoftware.foundation/en/insights/ green-software-is-a-priority-for-software-practitioners/, Accessed - 11/07/2024

    work page 2024

  7. [7]

    Towards a framework for carbon- aware virtual machine management,

    P. Pathania, R. Mehra, S. Sikand, V . S. Sharma, V . Kaulgud, N. Bamby, S. Podder, and A. P. Burden, “Towards a framework for carbon- aware virtual machine management,” in2024 IEEE 21st International Conference on Software Architecture Companion (ICSA-C). IEEE, 2024

    work page 2024

  8. [8]

    Mlca: a tool for machine learning life cycle assessment,

    C. Morand, A.-L. Ligozat, and A. N ´ev´eol, “Mlca: a tool for machine learning life cycle assessment,” in2024 10th International Conference on ICT for Sustainability (ICT4S). IEEE, 2024, pp. 227–238

    work page 2024

Show all 63 references
  1. [9]

    An experimental comparison of software-based power me- ters: focus on cpu and gpu,

    M. Jay, V . Ostapenco, L. Lef `evre, D. Trystram, A.-C. Orgerie, and B. Fichel, “An experimental comparison of software-based power me- ters: focus on cpu and gpu,” in2023 IEEE/ACM 23rd International Symposium on Cluster, Cloud and Internet Computing (CCGrid). IEEE, 2023

    2023

  2. [10]

    Towards sustainable ai: Monitoring and analysis of carbon emissions in machine learning algorithms,

    A. Martiny, “Towards sustainable ai: Monitoring and analysis of carbon emissions in machine learning algorithms,” Ph.D. dissertation, Politec- nico di Torino, 2023

    2023

  3. [11]

    Evaluating the carbon footprint of NLP methods: a survey and analysis of existing tools,

    N. Bannour, S. Ghannay, A. N ´ev´eol, and A.-L. Ligozat, “Evaluating the carbon footprint of NLP methods: a survey and analysis of existing tools,” inProceedings of the Second Workshop on Simple and Efficient Natural Language Processing. Association for Computational Linguis- ...

    2021

  4. [12]

    Software carbon intensity (sci) specification project,

    G. S. Foundation, “Software carbon intensity (sci) specification project,” Accessed - 13/07/2024. [Online]. Available: https://greensoftware. foundation/articles/software-carbon-intensity-sci-specification-project

    2024

  5. [13]

    Watts up? pro ac power meter for automated energy recording: A product review,

    J. M. Hirst, J. R. Miller, B. A. Kaplan, and D. D. Reed, “Watts up? pro ac power meter for automated energy recording: A product review,” Behavior Analysis in Practice, 2013

    2013

  6. [14]

    Kill a watt,

    p3international, “Kill a watt,” https://www.p3international.com/products/ p4400.html, Accessed - 01/07/2024

    2024

  7. [15]

    Powerblade: A low- profile, true-power, plug-through energy meter,

    S. DeBruin, B. Ghena, Y .-S. Kuo, and P. Dutta, “Powerblade: A low- profile, true-power, plug-through energy meter,” inProceedings of the 13th ACM Conference on Embedded Networked Sensor Systems, 2015

    2015

  8. [16]

    An energy metering and measurement soc with a novel low-power dsp architecture,

    B. Wu and N. Tan, “An energy metering and measurement soc with a novel low-power dsp architecture,”IEEE Internet of Things Journal, 2019

    2019

  9. [17]

    A review of energy measurement approaches,

    A. Noureddine, R. Rouvoy, and L. Seinturier, “A review of energy measurement approaches,”ACM SIGOPS Operating Systems Review, 2013

    2013

  10. [18]

    Rapl in action: Experiences in using rapl for power measurements,

    K. N. Khan, M. Hirki, T. Niemi, J. K. Nurminen, and Z. Ou, “Rapl in action: Experiences in using rapl for power measurements,”ACM Transactions on Modeling and Performance Evaluation of Computing Systems (TOMPECS), 2018

    2018

  11. [19]

    Turbostat(8),

    Linux.org, “Turbostat(8),” https://www.linux.org/docs/man8/turbostat. html, Accessed - 01/07/2024

    2024

  12. [20]

    Energy efficiency of personal computers: a comparative analysis,

    B. Prieto, J. J. Escobar, J. C. G ´omez-L´opez, A. F. D´ıaz, and T. Lampert, “Energy efficiency of personal computers: a comparative analysis,” Sustainability, 2022

    2022

  13. [21]

    PyPI, “pynvml,” https://pypi.org/project/nvidia-ml-py/, Accessed - 01/07/2024

    2024

  14. [22]

    Cloud jewels: Estimating kwh in the cloud,

    Etsy, “Cloud jewels: Estimating kwh in the cloud,” https://www.etsy. com/codeascraft/cloud-jewels-estimating-kwh-in-the-cloud/, Accessed - 01/07/2024

    2024

  15. [23]

    Thermal design power (tdp) in intel® processors,

    Intel, “Thermal design power (tdp) in intel® processors,” https://www.intel.com/content/www/us/en/support/articles/000055611/ processors.html, Accessed - 01/07/2024

    2024

  16. [24]

    Towards the systematic reporting of the energy and carbon footprints of machine learning,

    P. Henderson, J. Hu, J. Romoff, E. Brunskill, D. Jurafsky, and J. Pineau, “Towards the systematic reporting of the energy and carbon footprints of machine learning,”Journal of Machine Learning Research, 2020

    2020

  17. [25]

    Estimating aws ec2 instances power consumption,

    T. Benjamin Davy, “Estimating aws ec2 instances power consumption,” https://medium.com/teads-engineering/ estimating-aws-ec2-instances-power-consumption-c9745e347959, Accessed - 01/07/2024

    2024

  18. [26]

    Spec power®,

    S. P. E. Corporation, “Spec power®,” https://www.spec.org/power ssj2008/, Accessed - 01/07/2024

    2024

  19. [27]

    Computing server power modeling in a data center: Survey, taxonomy, and performance evaluation,

    L. Ismail and H. Materwala, “Computing server power modeling in a data center: Survey, taxonomy, and performance evaluation,”ACM Comput. Surv., 2020

    2020

  20. [28]

    How energy consumption in the cloud data center is calculated,

    S. DIOUANI and H. MEDROMI, “How energy consumption in the cloud data center is calculated,” in2019 International Conference of Computer Science and Renewable Energies (ICCSRE). IEEE, 2019

    2019

  21. [29]

    Energy consumption in cloud computing data centers,

    A. Uchechukwu, K. Li, Y . Shenet al., “Energy consumption in cloud computing data centers,”International Journal of Cloud Computing and Services Science (IJ-CLOSER), 2014

    2014

  22. [30]

    Llmcarbon: Modeling the end-to-end carbon footprint of large language models,

    A. Faiz, S. Kaneda, R. Wang, R. Osi, P. Sharma, F. Chen, and L. Jiang, “Llmcarbon: Modeling the end-to-end carbon footprint of large language models,”arXiv preprint arXiv:2309.14393, 2023

    2023 arXiv

  23. [31]

    Reducing the carbon impact of generative ai inference (today and in 2035),

    A. A. Chien, L. Lin, H. Nguyen, V . Rao, T. Sharma, and R. Wijayawar- dana, “Reducing the carbon impact of generative ai inference (today and in 2035),” inProceedings of the 2nd Workshop on Sustainable Computer Systems, ser. HotCarbon ’23. Association for Computing Machinery, 2023

    2023

  24. [32]

    Power provisioning for a warehouse-sized computer,

    X. Fan, W.-D. Weber, and L. A. Barroso, “Power provisioning for a warehouse-sized computer,” inProceedings of the 34th Annual Interna- tional Symposium on Computer Architecture, ser. ISCA ’07. Association for Computing Machinery, 2007

    2007

  25. [33]

    Esave: Estimating server and virtual machine energy,

    P. Pathania, R. Mehra, V . S. Sharma, V . Kaulgud, S. Podder, and A. P. Burden, “Esave: Estimating server and virtual machine energy,” inPro- ceedings of the 37th IEEE/ACM International Conference on Automated Software Engineering. Association for Computing Machinery, 2023

    2023

  26. [34]

    Customer carbon footprint tool,

    AWS, “Customer carbon footprint tool,” Accessed - 01/07/2024. [Online]. Available: https://aws.amazon.com/aws-cost-management/ aws-customer-carbon-footprint-tool/

    2024

  27. [35]

    Emissions impact dashboard,

    Microsoft, “Emissions impact dashboard,” https://www.microsoft. com/en-us/sustainability/emissions-impact-dashboard, Accessed - 01/07/2024

    2024

  28. [36]

    Carbon footprint,

    G. Cloud, “Carbon footprint,” https://cloud.google.com/carbon-footprint, Accessed - 01/07/2024

    2024

  29. [37]

    Charting the path towards sustain- able ai with azure machine learning resource metrics,

    M. Will Buchanan, “Charting the path towards sustain- able ai with azure machine learning resource metrics,” https://techcommunity.microsoft.com/t5/green-tech-blog/ charting-the-path-towards-sustainable-ai-with-azure-machine/ba-p/ 2866923, Accessed - 01/07/2024

    2024

  30. [38]

    Collect nvidia gpu metrics,

    AWS, “Collect nvidia gpu metrics,” https://docs. aws.amazon.com/AmazonCloudWatch/latest/monitoring/ CloudWatch-Agent-NVIDIA-GPU.html, Accessed - 01/07/2024

    2024

  31. [39]

    pyjoules 0.5.1,

    PyPI, “pyjoules 0.5.1,” https://pypi.org/project/pyJoules/, Accessed - 01/07/2024

    2024

  32. [40]

    nvmldevicegetpowerusage,

    N. D. Zone, “nvmldevicegetpowerusage,” https://developer.nvidia.com/ management-library-nvml, Accessed - 01/07/2024

    2024

  33. [41]

    Codecarbon,

    CodeCarbon, “Codecarbon,” https://codecarbon.io/, Accessed - 01/07/2024

    2024

  34. [42]

    Eco2ai: carbon emissions tracking of machine learning models as the first step towards sustainable ai,

    S. A. Budennyy, V . D. Lazarev, N. N. Zakharenko, A. N. Korovin, O. Plosskaya, D. V . Dimitrov, V . Akhripkin, I. Pavlov, I. V . Oseledets, I. S. Barsolaet al., “Eco2ai: carbon emissions tracking of machine learning models as the first step towards sustainable ai,” inDoklady M...

    2022

  35. [43]

    Mlco2 impact,

    M. Impact, “Mlco2 impact,” https://mlco2.github.io/impact/, Accessed - 13/07/2024

    2024

  36. [44]

    Carbontracker: Tracking and predicting the carbon footprint of training deep learning models,

    L. F. W. Anthony, B. Kanding, and R. Selvan, “Carbontracker: Tracking and predicting the carbon footprint of training deep learning models,” arXiv preprint arXiv:2007.03051, 2020

    2007 arXiv

  37. [45]

    Cloud carbon footprint,

    C. C. Footprint, “Cloud carbon footprint,” https://www. cloudcarbonfootprint.org/, Accessed - 01/07/2024

    2024

  38. [46]

    Lean ict,

    T. S. Project, “Lean ict,” https://theshiftproject.org/en/lean-ict-2/, Ac- cessed - 01/07/2024

    2024

  39. [47]

    Cumulator—a tool to quantify and report the carbon foot- print of machine learning computations and communication in academia and healthcare,

    T. Tr ´ebaol, “Cumulator—a tool to quantify and report the carbon foot- print of machine learning computations and communication in academia and healthcare,” 2020

    2020

  40. [48]

    Tracarbon,

    F. Valeye, “Tracarbon,” https://github.com/fvaleye/tracarbon, Accessed - 01/07/2024

    2024

  41. [49]

    Green algorithms: quanti- fying the carbon footprint of computation,

    L. Lannelongue, J. Grealey, and M. Inouye, “Green algorithms: quanti- fying the carbon footprint of computation,”Advanced science, 2021

    2021

  42. [50]

    Powerjoular and joularjx: Multi-platform software power monitoring tools,

    A. Noureddine, “Powerjoular and joularjx: Multi-platform software power monitoring tools,” in2022 18th International Conference on Intelligent Environments (IE). IEEE, 2022

    2022

  43. [51]

    Likwid: A lightweight performance-oriented tool suite for x86 multicore environments,

    J. Treibig, G. Hager, and G. Wellein, “Likwid: A lightweight performance-oriented tool suite for x86 multicore environments,” in 2010 39th international conference on parallel processing workshops. IEEE, 2010

    2010

  44. [52]

    Pmt: Power measurement toolkit,

    S. Corda, B. Veenboer, and E. Tolley, “Pmt: Power measurement toolkit,” in2022 IEEE/ACM International Workshop on HPC User Support Tools (HUST). IEEE, 2022, pp. 44–47

    2022

  45. [53]

    Rocm system management interface (rocm smi) library,

    AMD, “Rocm system management interface (rocm smi) library,” https://rocm.docs.amd.com/projects/rocm smi lib/en/latest/, Accessed - 01/07/2024

    2024

  46. [54]

    Smartwatts: Self-calibrating software-defined power meter for containers,

    G. Fieni, R. Rouvoy, and L. Seinturier, “Smartwatts: Self-calibrating software-defined power meter for containers,” in2020 20th IEEE/ACM International Symposium on Cluster, Cloud and Internet Computing (CCGRID). IEEE, 2020

    2020

  47. [55]

    Carbon footprint,

    C. Footprint, “Carbon footprint,” https://www.carbonfootprint.com/, Ac- cessed - 15/07/2024

    2024

  48. [56]

    Co2 signal,

    C. signal, “Co2 signal,” https://www.co2signal.com/, Accessed - 11/07/2024

    2024

  49. [57]

    Electricity maps api,

    E. Maps, “Electricity maps api,” https://static.electricitymaps.com/api/ docs/index.html, Accessed - 01/07/2024

    2024

  50. [58]

    Real-time data can help to track and reduce emissions from electricity sector,

    IEA.org, “Real-time data can help to track and reduce emissions from electricity sector,” Accessed - 29/01/2025. [Online]. Available: https://www.iea.org/commentaries/ real-time-data-can-help-to-track-and-reduce-emissions-from-electricity-sector? utm source=chatgpt.com

    2025

  51. [59]

    Embodied emissions,

    C. C. Footprint, “Embodied emissions,” https://www. cloudcarbonfootprint.org/docs/embodied-emissions/, Accessed - 12/07/2024

    2024

  52. [60]

    Act: designing sustainable computer systems with an architectural carbon modeling tool,

    U. Gupta, M. Elgamal, G. Hills, G.-Y . Wei, H.-H. S. Lee, D. Brooks, and C.-J. Wu, “Act: designing sustainable computer systems with an architectural carbon modeling tool,” inProceedings of the 49th Annual International Symposium on Computer Architecture, ser. ISCA ’22. Associ...

    2022

  53. [61]

    Measuring processor power tdp vs.acp,

    Intel, “Measuring processor power tdp vs.acp,” https://www.intel.com/content/dam/doc/white-paper/ resources-xeon-measuring-processor-power-paper.pdf, Accessed - 12/07/2024

    2024

  54. [62]

    Software carbon intensity (sci) spec- ification achieves iso standard status, advancing green software development,

    G. S. Foundation, “Software carbon intensity (sci) spec- ification achieves iso standard status, advancing green software development,” https://greensoftware.foundation/articles/ sci-specification-achieves-iso-standard-status, Accessed - 15/07/2024

    2024

  55. [63]

    Comet integration,

    CodeCarbon, “Comet integration,” https://mlco2.github.io/codecarbon/ comet.html, Accessed - 13/07/2024

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.