Pith. sign in

REVIEW 3 major objections 5 minor 8 cited by

Life-Cycle Emissions of AI Hardware: A Cradle-To-Grave Approach and Generational Trends

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read This paper provides the first cradle-to-grave carbon ledger for AI accelerators and a metric that shows a threefold gain across two generations.

desk verdict First real manufacturing emissions for an AI accelerator plus a fleet-measured carbon metric—genuinely useful, but the headline 3x rests on a one-month snapshot and needs uncertainty bounds before it's a fact. read the letter →

arxiv 2502.01671 v1 pith:3LTU3ED4 submitted 2025-02-01 cs.AR cs.AI

Ian Schneider , Hui Xu , Stephan Benecke , David Patterson , Keguo Huang , Parthasarathy Ranganathan , Cooper Elsworth This is my paper · ORCID
classification cs.ARcs.AI
keywords ArtificialIntelligenceTPUAIAcceleratorCarbonAccountingLife-CycleAnalysisSustainableComputeIntensityGreenhouseGasEmissions
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

This paper aims to establish the first complete cradle-to-grave accounting of greenhouse gas emissions for AI accelerator hardware, including the first published estimate of the emissions embedded in manufacturing an AI accelerator. Using five generations of tensor processing units (TPUs), it reports that operational electricity dominates lifetime emissions, that manufacturing emissions grow with each generation but are outpaced by performance gains, and that its new metric, compute carbon intensity (CCI), improved threefold from the TPU v4i to the TPU v6e. The wider point is that this LCA is deliberately written as a recipe, so other hardware designers can produce comparable numbers for their own chips.

What carries the argument

The engine of the argument is compute carbon intensity (CCI), defined as grams of $\mathrm{CO_2e}$ per exaFLOP of utilized floating-point operations, with the FLOP count taken from runtime counters on deployed machines. CCI splits into embodied and operational parts, and operational CCI obeys the relation $\text{Operational CCI} = \frac{\text{electricity emissions factor}}{\text{utilized FLOPs per joule}}$. The measurement machinery is a fleet snapshot: five-minute power readings from each tray's power supply, a count of utilized FLOPs per chip per five-minute interval, a 1.10 power usage effectiveness multiplier, and propensity-score weighting that equalizes duty cycles across generations before CCI is computed. The functional unit is one AI computer---accelerator trays plus host tray---over a six-year lifetime, following the greenhouse-gas protocol's scope definitions and standard LCA practice.

What would settle it

Recompute lifetime CCI for the same TPU generations from a full year of fleet power and utilization measurements instead of the one-month snapshot from October 2024; if yearly average duty cycle, power draw, or workload mix differs, the reported 3x improvement from TPU v4i to TPU v6e will change.

Watch

Extended reading notes

Core claim

The central discovery is that the total carbon story of an AI machine can be compressed into a single number---CCI, measured as grams of $\mathrm{CO_2e}$ per exaFLOP of utilized computation---and that for the five TPU generations studied this number falls with each generation. Fleet-wide measurements of power and utilized floating-point operations, adjusted with propensity-score weighting to remove utilization differences, show CCI improving 3x from TPU v4i to TPU v6e. The paper also reports the underlying absolutes: operational emissions are 70\textendash 90% of lifetime emissions depending on accounting method, embodied emissions from manufacturing, transport, and construction run from roughly 390 to 1,100 kg $\mathrm{CO_2e}$ per machine, and manufacturing emissions rise about 1.8x from v4i to v6e while peak performance rises about 4.7x. On the paper's account, the performance gain of newer chips more than compensates for the extra carbon embedded in their manufacture.

Load-bearing premise

The lifetime numbers come from multiplying one month of measured power and utilization by an assumed six-year lifespan, so if machines are used more or less as they age, or if grid carbon intensity shifts, the reported lifetime emissions and the 3x improvement would move.

Editorial extensions

If this is right

  • With CCI, a team can convert a model's FLOP count into a ballpark carbon footprint by multiplying by the hardware's CCI; the paper illustrates this by estimating about 107 tonnes of $\mathrm{CO_2e}$ for GPT-3 on one TPU generation and 89 tonnes on the next.
  • Because operational electricity is 70\textendash 90% of lifetime emissions depending on accounting, energy efficiency and clean-power procurement are the largest levers now, and embodied emissions grow in relative importance as grids decarbonize.
  • Manufacturing emissions rise from one TPU generation to the next, yet embodied CCI falls, so the added carbon in bigger dies, HBM, and DRAM is outweighed by the added computation those parts deliver.
  • Under hourly 24/7 clean-energy accounting, a v6e with 90% local clean power would cut lifetime electricity emissions about 3.3x, and adding clean manufacturing yields a 10\textendash 14x CCI improvement over a v4i.
  • The appendices give a step-by-step LCA recipe; if other vendors follow it, the industry can compare accelerators on carbon per unit of computation the way it compares them on cost.

Reading between the lines

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

  • The one-month snapshot makes CCI a point-in-time measure; extending the measurement window to a full year would show whether seasonal grid carbon and workload mix change the reported 3x improvement.
  • The component-level emissions breakdown points to a design rule the paper does not spell out: die area, HBM capacity, and host DRAM are the carbon knobs that dominate embodied emissions, so memory is the part to watch as chips scale.
  • Because CCI's denominator is the actual fleet FLOP mix, the metric is workload-dependent; a standardized benchmark version of CCI would let different vendors publish comparable numbers without revealing fleet utilization.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript presents a cradle-to-grave life-cycle assessment (LCA) of five Google Tensor Processing Unit generations (TPU v4i, v5e, v6e, v4, v5p), using first-party data for manufacturing, operational, and end-of-life stages. It introduces a new normalized metric, compute carbon intensity (CCI, in gCO2e per ExaFLOP), and reports that CCI improves 3x from TPU v4i to TPU v6e. The paper also provides breakdowns of embodied versus operational emissions, contrasts LCA results with corporate inventory accounting, and includes tutorial-style appendices describing the LCA process.

Significance. If the results hold, the paper is potentially valuable as the first public cradle-to-grave LCA of AI accelerators and the first publication of manufacturing emissions for an AI accelerator. The CCI metric, based on fleet-measured utilized FLOPs and power rather than TDP, is a useful normalization for comparing AI hardware sustainability. The detailed appendices, explicit inventory boundary, and comparison with existing accounting frameworks are strengths. However, the headline quantitative claims rest on a single-month operational snapshot and on point estimates without uncertainty quantification; these issues must be addressed before the significance can be fully realized.

major comments (3)
  1. [Section 4.1 and Appendix A.3] The headline claim that CCI improves 3x from TPU v4i to TPU v6e is computed from a one-month fleet snapshot (October 2024) linearly extrapolated to an assumed six-year lifetime, with v6e data collected before its December 2024 customer launch. Different generations are observed at different points in their life cycles (e.g., v4i four years after deployment, v6e pre-launch), so any age-dependent trends in power, utilization, or workload mix would bias the comparison. The paper reports no fleet sizes, no month-to-month variability, and no sensitivity analysis for the six-year lifetime assumption. As stated in Section 1, the operational measurement is explicitly a 'snapshot in time'; the lifetime CCI interpretation therefore needs additional support or a carefully qualified framing.
  2. [Appendix F] Propensity score weighting balances duty-cycle distributions, but it cannot correct for the key confounders identified above: the mix of active workloads (compute-dense versus memory-bound), deployment maturity, and the fact that the snapshot occurs at different points in each generation's life. The weighting procedure is described at a high level, but Table 4 reports only values normalized to the cohort baseline (v4i or v4 = 1), not the actual duty cycle means, standard deviations, or observation counts. Without these details, the reader cannot assess how well the weighting works or whether the residual imbalance affects the 3x claim.
  3. [Table 1 and Sections 2, 4.3, 5] All LCA results are point estimates with no uncertainty bounds. The operational numbers depend on a single PUE (1.10), a fixed six-year lifetime, and one-year average emission factors; the manufacturing numbers depend on proprietary IMEC virtual fab parameters, yield rates, and abatement ratios. Given the multiplicity of assumptions in this study, the quantitative comparisons (e.g., 3x, 10x, 14x improvements) need at least a one-way sensitivity analysis or upper/lower bounds to establish robustness. Without this, it is unclear whether the generational ordering and magnitudes are statistically or practically significant.
minor comments (5)
  1. [Table 4] Table 4 is confusing because all duty-cycle and performance values are normalized to the cohort baseline; please report the actual duty-cycle means, standard deviations, and observation counts for each generation before and after weighting.
  2. [References] References [52] and [53] are the same paper (Vahdat, Ma, and Patterson, 'New Computer Evaluation Metrics for a Changing World') and should be merged into a single entry.
  3. [Section 4.1] The sentence 'The quadrupling of its systolic array size explains in part v6e's large gain' is unsupported by a citation or a derived calculation; please provide a reference or a caveat.
  4. [Appendix F] The propensity score model is described without specifying the number of duty-cycle levels, the level boundaries, or whether any covariates beyond duty cycle are included; please provide these details for reproducibility.
  5. [Introduction] The claim of being 'the first publication of manufacturing emissions of an AI accelerator' is strong; consider softening to 'to the best of our knowledge' and explicitly discuss how the prior server LCAs reviewed in [22] overlap or differ in scope.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the CCI values are direct fleet measurements combined with a bottom-up LCA, and no headline result is fitted to reproduce itself. The only self-citations are minor and not load-bearing.

full rationale

The paper's central derivation chain is empirical rather than definitional. CCI is computed as lifetime CO2e divided by measured utilized FLOPs, where operational CO2e comes from first-party PSU power readings and five-minute utilized-FLOP counters across Google's fleet (Appendix A.3), and embodied CO2e comes from a bottom-up LCA using IMEC.netzero and proprietary manufacturing parameters (die size, technology node, yields, abatement ratios). No parameter is fitted to reproduce the 3x v4i-to-v6e improvement; indeed, propensity-score weighting makes the raw improvement more modest (the CCI reduction changes from 74% to 66%, Table 4), and the 3x headline is the resulting weighted ratio. The GPT-3 estimate is a straightforward multiplication of the measured Table 1 CCI by the externally known 3.14e23 FLOP count, so it is an application of the metric, not a self-referential validation. The embodied-emissions results are outputs of an LCA model, not a rename of prior published values, and the paper explicitly compares them to external proxies rather than importing them. The only self-citations of note are Vahdat et al. [53] for the CO2e/Goodput inspiration and the six-year lifetime assumption [52]; neither is load-bearing for the main result, because the six-year lifetime cancels out of operational CCI and is applied uniformly across generations, so it cannot force the 3x comparison. The one-month fleet snapshot extrapolated to a six-year lifetime is a legitimate measurement-representativeness concern, but it is not circularity: it is an empirical validity limitation, not a reduction of the result to its own inputs.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The central claim rests on measured fleet data (power, FLOPs, duty cycle) and on modeled manufacturing inventories. Key modeling choices: six-year lifespan, 1.10 PUE, Google-wide 2023 emission factors (LB 366, MB 135, 24/7 212 gCO2e/kWh), and the IMEC virtual fab parameterization. These are disclosed but not independently verifiable from the paper. No free parameters are fitted to force the 3x result; the improvement is computed directly from the chosen inputs.

free parameters (3)
  • Machine lifespan = 6 years
    Assumed lifetime for amortizing embodied and operational emissions; directly scales all lifetime totals and CCI.
  • Data center PUE = 1.10
    Google 2024 average power usage effectiveness applied to operational energy; from Google's environmental report [12].
  • Electricity emission factors (LB, MB, 24/7) = 366, 135, 212 gCO2e/kWh
    Google 2023 fleet-wide factors used to convert operational kWh into CO2e; MB reflects CFE procurement credits.
assumptions (4)
  • domain assumption IMEC.netzero virtual fab and Google's proprietary LCIs accurately model TPU manufacturing emissions when parameterized with chip data.
    Appendix A.2 relies on these models; errors in wafer or packaging emission factors propagate into all embodied numbers.
  • domain assumption Duty cycle is the only significant confounder between TPU generations.
    Propensity score weighting in Appendix F adjusts only for duty cycle; workload mix, data center location, and software versions could also differ across generations.
  • domain assumption Runtime FLOP counters accurately measure utilized computation.
    CCI's denominator is defined from these counters (Section 3); counter limits or mixed numeric formats affect the metric.
  • standard math IPCC AR5 GWP100 values are the correct characterization factors.
    Used to convert non-CO2 GHGs to CO2e; standard in the field but a choice that affects totals.
invented entities (1)
  • Compute carbon intensity (CCI) independent evidence
    purpose: New metric normalizing lifecycle CO2e per FLOP to compare AI hardware across generations and estimate workload footprints.
    Defined in Section 3 from measured power, FLOP counters, and emission factors; other groups with access to their own telemetry can compute the same quantity, giving it an independent falsifiable handle.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Life-Cycle Emissions of AI Hardware: A Cradle-To-Grave Approach and Generational Trends." pith.science (2026). https://pith.science/paper/3LTU3ED4

@misc{pith2026250201671,
  author       = {Pith},
  title        = {Pith review of: Life-Cycle Emissions of AI Hardware: A Cradle-To-Grave Approach and Generational Trends},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3LTU3ED4}},
  note         = {Machine review of arXiv:2502.01671}
}
read the original abstract

Specialized hardware accelerators aid the rapid advancement of artificial intelligence (AI), and their efficiency impacts AI's environmental sustainability. This study presents the first publication of a comprehensive AI accelerator life-cycle assessment (LCA) of greenhouse gas emissions, including the first publication of manufacturing emissions of an AI accelerator. Our analysis of five Tensor Processing Units (TPUs) encompasses all stages of the hardware lifespan - from raw material extraction, manufacturing, and disposal, to energy consumption during development, deployment, and serving of AI models. Using first-party data, it offers the most comprehensive evaluation to date of AI hardware's environmental impact. We include detailed descriptions of our LCA to act as a tutorial, road map, and inspiration for other computer engineers to perform similar LCAs to help us all understand the environmental impacts of our chips and of AI. A byproduct of this study is the new metric compute carbon intensity (CCI) that is helpful in evaluating AI hardware sustainability and in estimating the carbon footprint of training and inference. This study shows that CCI improves 3x from TPU v4i to TPU v6e. Moreover, while this paper's focus is on hardware, software advancements leverage and amplify these gains.

Figures

Figures reproduced from arXiv: 2502.01671 by the authors.

Figure 1
Figure 1. A comprehensive AI cradle-to-grave emissions in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Lifetime AI hardware compute carbon intensity [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Accelerator and host emissions by life-cycle stage. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Absolute manufacturing emissions per chip, by generation and by component type, and average TeraFLOP/second. It [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Typical hardware components in a server. [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: A generic schematic cross-sectional sketch of a TPU [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Schematic of the process flow for the manufacturing [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Duty Cycle and Power over time for RLHF on v5e [PITH_FULL_IMAGE:figures/full_fig_p014_8.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 8 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Lights, Camera, Carbon: Architectural Scaling Laws for Video Generation Energy Consumption

    cs.MM 2026-07 conditional novelty 6.5 of 10

    Energy of text-to-video diffusion models is predicted from architectural first principles and observable generation parameters with under 3% MAPE, without needing weights or model size.

  2. CarbonFlex: Enabling Carbon-aware Provisioning and Scheduling for Cloud Clusters

    cs.DC 2025-05 conditional novelty 6.5 of 10

    A learning-based cluster scheduler that jointly provisions capacity and elastically scales batch jobs to cut operational carbon by up to 57.5% in AWS tests.

  3. Auto-Scaling Heterogeneous Neural Processing Units for Energy and Cost-Efficient LLM Serving

    cs.AR 2026-07 conditional novelty 6.0 of 10

    NeuScale routes LLM inference requests to the most energy/cost-efficient configuration of heterogeneous NPU chips using roofline allocation and runtime auto-scaling.

  4. CEO-DC: Driving Decarbonization in HPC Data Centers with Actionable Insights

    cs.AR 2025-07 conditional novelty 6.0 of 10

    A decision framework using new carbon and price efficiency metrics shows most AI platform improvements cannot keep pace with demand growth, and short upgrade cycles need carbon prices far above current levels.

  5. Misinformation by Omission: The Need for More Environmental Transparency in AI

    cs.CY 2025-06 conditional novelty 6.0 of 10

    Environmental disclosure for notable AI models peaked in 2022 and then declined, and out-of-context energy and emissions estimates now dominate media coverage.

  6. EcoServe: Designing Carbon-Aware AI Inference Systems

    cs.DC 2025-02 conditional novelty 6.0 of 10

    EcoServe combines four strategies (reuse, rightsize, reduce, recycle) in an ILP optimizer to cut modeled carbon emissions for LLM serving by up to 47% while keeping SLOs.

  7. LLM Harms: A Taxonomy and Discussion

    cs.CY 2025-12 unverdicted novelty 3.0 of 10

    This paper proposes a taxonomy of LLM harms in five categories and suggests mitigation strategies plus a dynamic auditing system for responsible development.

  8. A Survey of End-to-End Modeling for Distributed DNN Training: Workloads, Simulators, and TCO

    cs.DC 2025-06 conditional novelty 2.0 of 10

    This survey classifies distributed DNN training simulators into analytical, profiling-based, and execution-driven categories, and compares them alongside TCO and carbon-emission models.

Reference graph

Works this paper leans on

58 extracted references · 41 canonical work pages · cited by 8 Pith papers

  1. [1]

    International Energy Agency. 2022. Advancing Decarbonisation through Clean Electricity Procurement. https://www.iea.org/reports/advancing- decarbonisation-through-clean-electricity-procurement

    work page 2022

  2. [2]

    Apple. 2019. Product environmental report Mac Pro. https://www.apple.com/ environment/pdf/products/desktops/Mac_Pro_PER_Dec2019.pdf

    work page 2019

  3. [3]

    Anders Bjørn, Shannon M Lloyd, Matthew Brander, and H Damon Matthews. 2022. Renewable energy certificates threaten the integrity of corporate science-based targets. Nature Climate Change 12, 6 (2022), 539–546

    work page 2022

  4. [4]

    Lucía Bouza, Aurélie Bugeau, and Loïc Lannelongue. 2023. How to estimate carbon footprint when training deep learning models? A guide and review. Envi- ronmental Research Communications 5, 11 (2023), 115014

    work page 2023

  5. [5]

    Google Cloud. 2024. Carbon Footprint reporting methodology. https://cloud. google.com/carbon-footprint/docs/methodology

    work page 2024

  6. [6]

    Benjamin Davy. 2021. Building an AWS EC2 carbon emissions dataset. https://medium.com/teads-engineering/building-an-aws-ec2-carbon- emissions-dataset-3f0fd76c98ac

    work page 2021

  7. [7]

    Jacques A de Chalendar and Sally M Benson. 2019. Why 100% renewable energy is not enough. Joule 3, 6 (2019), 1389–1393

    work page 2019

  8. [8]

    Xiaobo Fan, Wolf-Dietrich Weber, and Luiz Andre Barroso. 2007. Power pro- visioning for a warehouse-sized computer. In Proc. 34th Annual International Symposium on Computer Architecture (ISCA) (San Diego, CA, USA). 13–23

    work page 2007

Show all 58 references
  1. [9]

    Rolf Frischknecht, Niels Jungbluth, Hans-Jörg Althaus, Gabor Doka, Roberto Dones, Thomas Heck, Stefanie Hellweg, Roland Hischier, Thomas Nemecek, Gerald Rebitzer, et al. 2005. The ecoinvent database: overview and methodological framework. The international journal of life cycl...

    2005

  2. [10]

    Google. 2018. Moving toward 24x7 Carbon-Free Energy at Google Data Centers: Progress and Insights. Google Report. https://www.gstatic.com/gumdrop/ sustainability/24x7-carbon-free-energy-data-centers.pdf

    2018

  3. [11]

    Google. 2024. The Corporate Role in Accelerating Advanced Clean Electricity Technologies. Google Report. https://www.gstatic.com/gumdrop/sustainability/ accelerating-advanced-clean-electricity-technologies.pdf

    2024

  4. [12]

    Google. 2024. Google 2024 Environmental Report . Technical Report. Google

    2024

  5. [13]

    Google. 2024. Google Gemini AI update, December 2024. Google Blog. https://blog.google/technology/google-deepmind/google-gemini-ai- update-december-2024/#gemini-2-0-flash

    2024

  6. [14]

    Greenhouse Gas Protocol. 2004. The Greenhouse Gas Protocol: A Corporate Ac- counting and Reporting Standard . World Resources Institute and World Business Council for Sustainable Development

    2004

  7. [15]

    Udit Gupta, Mariam Elgamal, Gage Hills, Gu-Yeon Wei, Hsien-Hsin S Lee, David Brooks, and Carole-Jean Wu. 2022. ACT: Designing sustainable computer systems with an architectural carbon modeling tool. In Proc. of 49th Annual International Symposium on Computer Architecture (ISCA...

    2022

  8. [16]

    Udit Gupta, Young Geun Kim, Sylvia Lee, Jordan Tse, Hsien-Hsin S Lee, Gu-Yeon Wei, David Brooks, and Carole-Jean Wu. 2021. Chasing carbon: The elusive environmental footprint of computing. In 2021 IEEE International Symposium on High-Performance Computer Architecture (HPCA). 854–867

    2021

  9. [17]

    Danny Hernandez and Tom B Brown. 2020. Measuring the algorithmic efficiency of neural networks. arXiv preprint arXiv:2005.04305 (2020)

    2020 arXiv

  10. [18]

    Anson Ho, Tamay Besiroglu, Ege Erdil, David Owen, Robi Rahman, Zifan Carl Guo, David Atkinson, Neil Thompson, and Jaime Sevilla. 2024. Algorithmic progress in language models. arXiv preprint arXiv:2403.05812 (2024)

    2024 arXiv

  11. [19]

    John Houghton, G. J. Jenkins, and J. J. Ephraums (Eds.). 1990. Climate Change: The IPCC Scientific Assessment . Cambridge University Press

    1990

  12. [20]

    IMEC. 2024. IMEC.netzero virtual fab. https://netzero.imec-int.com/

    2024

  13. [21]

    International Energy Agency (IEA). 2024. What the data centre and AI boom could mean for the energy sector. https://www.iea.org/commentaries/what-the- data-centre-and-ai-boom-could-mean-for-the-energy-sector. Licence: CC BY 4.0

    2024

  14. [22]

    Shixin Ji, Zhuoping Yang, Xingzhen Chen, Stephen Cahoon, Jingtong Hu, Yiyu Shi, Alex K Jones, and Peipei Zhou. 2024. SCARIF: Towards Carbon Modeling of Cloud Servers with Accelerators. In 2024 IEEE Computer Society Annual Symposium on VLSI (ISVLSI). 496–501

    2024

  15. [23]

    Jouppi, Doe Hyun Yoon, Matthew Ashcraft, Mark Gottscho, Thomas B

    Norman P. Jouppi, Doe Hyun Yoon, Matthew Ashcraft, Mark Gottscho, Thomas B. Jablin, George Kurian, James Laudon, Sheng Li, Peter Ma, Xiaoyu Ma, Thomas Norrie, Nishant Patil, Sushma Prasad, Cliff Young, Zongwei Zhou, and David Patterson. 2021. Ten Lessons From Three Generations...

    2021

  16. [24]

    Norman P. Jouppi, George Kurian, Sheng Li, Peter Ma, Rahul Nagarajan, Lifeng Nai, Nishant Patil, Suvinay Subramanian, Andy Swing, Brian Towles, Clifford Young, Xiang Zhou, Zongwei Zhou, and David Patterson. 2023. TPU v4: An Optically Reconfigurable Supercomputer for Machine Le...

    2023 doi

  17. [25]

    Norman P Jouppi, Cliff Young, Nishant Patil, David Patterson, Gaurav Agrawal, Raminder Bajwa, Sarah Bates, Suresh Bhatia, Nan Boden, Al Borchers, et al. 2017. In-datacenter performance analysis of a tensor processing unit. In Proceedings of the 44th annual international sympos...

    2017

  18. [26]

    Walter Klöpffer and Birgit Grahl. 2014. Life cycle assessment (LCA): a guide to best practice. John Wiley & Sons

    2014

  19. [27]

    Tsai-Chi Kuo, Chien-Yun Kuo, and Liang-Wei Chen. 2022. Assessing environ- mental impacts of nanoscale semi-conductor manufacturing from the life cy- cle assessment perspective. Resources, Conservation and Recycling 182 (2022). https://doi.org/10.1016/j.resconrec.2022.106289

    2022

  20. [28]

    Loïc Lannelongue, Jason Grealey, and Michael Inouye. 2021. Green algorithms: quantifying the carbon footprint of computation. Advanced science 8, 12 (2021), 2100707

    2021

  21. [29]

    Melissa C Lott, Abraham Silverman, Qëndresa Krasniqi, Harry Ken- nard, and Jackie Ratner. 2023. Rethinking the Greenhouse Gas Pro- tocol: Insights from an Expert Stakeholder Engagement. (2023). https://www.energypolicy.columbia.edu/publications/revising-the- greenhouse-gas-pro...

    2023

  22. [30]

    Alexandra Sasha Luccioni and Alex Hernandez-Garcia. 2023. Counting Car- bon: A Survey of Factors Influencing the Emissions of Machine Learning. arXiv:2302.08476 https://arxiv.org/abs/2302.08476

    2023 arXiv

  23. [31]

    Alexandra Sasha Luccioni, Sylvain Viguier, and Anne-Laure Ligozat. 2023. Es- timating the Carbon Footprint of BLOOM, a 176B Parameter Language Model. Journal of Machine Learning Research 24 (2023), 1–15. https://doi.org/10.48550/ arXiv.2211.02001

    2023 doi arXiv

  24. [32]

    Gunnar Myhre, Drew Shindell, F-M Bréon, William Collins, Jan Fuglestvedt, Jianping Huang, Dorothy Koch, J-F Lamarque, David Lee, Blanca Mendoza, et al

  25. [33]

    United Nations. 2021. 24/7 Carbon-free Energy Compact. https://www.un.org/ en/energy-compacts/page/compact-247-carbon-free-energy

    2021

  26. [34]

    Dylan Patel and Aleksandar Kostovic. 2023. TPUv5e: The New Benchmark in Cost-Efficient Inference and Training for <200B Parameter Models: Latency, Performance, Fine-tuning, Scaling, and Networking. Semianalysis Report. https: //semianalysis.com/2023/09/01/tpuv5e-the-new-benchm...

    2023

  27. [35]

    David Patterson, Jeffrey M Gilbert, Marco Gruteser, Efren Robles, Krishna Sekar, Yong Wei, and Tenghui Zhu. 2024. Energy and Emissions of Machine Learning on Smartphones vs. the Cloud. Commun. ACM 67, 2 (2024), 86–97

    2024

  28. [36]

    David Patterson, Joseph Gonzalez, Urs Hölzle, Quoc Le, Chen Liang, Lluis-Miquel Munguia, Daniel Rothchild, David R So, Maud Texier, and Jeff Dean. 2022. The Carbon Footprint of Machine Learning Training Will Plateau, Then Shrink. IEEE Computer 55, 7 (2022), 18–28. https://doi....

    2022

  29. [37]

    David Patterson, Joseph Gonzalez, Quoc Le, Chen Liang, Lluis-Miquel Munguia, Daniel Rothchild, David So, Maud Texier, and Jeff Dean. 2021. Carbon Emissions and Large Neural Network Training. https://arxiv.org/abs/2104.10350

    2021 arXiv

  30. [38]

    Adele Peters. 2024. Amazon says it hit a goal of 100% clean power. Employees say it’s more like 22%. https://www.fastcompany.com/91153918/amazon-says- 15 arXiv, February, 2025 Ian Schneider, Hui Xu, Stephan Benecke, David Patterson, Keguo Huang, Parthasarathy Ranganathan, and ...

    2024

  31. [39]

    Robi Rahman, David Owen, and Josh You. 2024. Tracking Large-Scale AI Models. https://epoch.ai/blog/tracking-large-scale-ai-models

    2024

  32. [40]

    Iegor Riepin and Tom Brown. 2024. On the means, costs, and system-level impacts of 24/7 carbon-free energy procurement. Energy Strategy Reviews 54 (2024), 101488

    2024

  33. [41]

    Paul Rosenbaum and Donald Rubin. 1983. The Central Role of the Propensity Score in Observational Studies For Causal Effects. Biometrika 70 (04 1983), 41–55. https://doi.org/10.1093/biomet/70.1.41

    1983 doi

  34. [42]

    Ian Schneider and Taylor Mattia. 2024. Carbon accounting in the Cloud: a methodology for allocating emissions across data center users. arXiv:2406.09645 https://arxiv.org/abs/2406.09645

    2024 arXiv

  35. [43]

    Roy Schwartz, Jesse Dodge, Noah A Smith, and Oren Etzioni. 2020. Green AI. Commun. ACM 63, 12 (2020), 54–63

    2020

  36. [44]

    Mary Elizabeth Sotos. 2015. GHG protocol scope 2 guidance. (2015)

    2015

  37. [45]

    Sphera. 2024. LCA for Experts . https://sphera.com/solutions/product- stewardship/life-cycle-assessment-software-and-data/lca-for-experts/

    2024

  38. [46]

    Christoforos Spiliotopoulos, David Bernad-Beltrán, and Leonidas Milios. 2024. A methodological proposal for determining priority product categories towards the development of a reparability score in the European Union. In 2024 Elec- tronics Goes Green 2024+(EGG) . IEEE, 1–6. h...

    2024

  39. [47]

    Emma Strubell, Ananya Ganesh, and Andrew McCallum. 2019. Energy and Policy Considerations for Deep Learning in NLP. arXiv:1906.02243 https: //arxiv.org/abs/1906.02243

    2019 arXiv

  40. [48]

    Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yas- mine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhos- ale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288 (2023)

    2023 arXiv

  41. [49]

    Tristan Trébaol. 2020. CUMULATOR—a tool to quantify and report the carbon footprint of machine learning computations and communication in academia and healthcare

    2020

  42. [50]

    TSMC. 2023. 2023 Sustainability Report. https://esg.tsmc.com/en-US/file/public/ e-all_2023.pdf

    2023

  43. [51]

    Amin Vahdat and Mark Lohmeyer. 2023. Enabling next-generation AI workloads: Announcing TPU v5p and AI Hypercomputer. https: //cloud.google.com/blog/products/ai-machine-learning/introducing-cloud-tpu- v5p-and-ai-hypercomputer

    2023

  44. [52]

    Amin Vahdat, Xiaoyu Ma, and David Patterson. 2024. New Computer Evaluation Metrics for a Changing World. Commun. ACM 67, 10 (Sept. 2024), 31–33. https: //doi.org/10.1145/3637867

    2024 doi

  45. [53]

    Amin Vahdat, Xiaoyu Ma, and David Patterson. 2024. New Computer Evaluation Metrics for a Changing World. Commun. ACM 67, 10 (2024), 31–33

    2024

  46. [54]

    Jaylen Wang, Daniel S. Berger, Fiodar Kazhamiaka, Celine Irvene, Chaojie Zhang, Esha Choukse, Kali Frost, Rodrigo Fonseca, Brijesh Warrier, Chetan Bansal, Jonathan Stern, Ricardo Bianchini, and Akshitha Sriraman. 2024. Designing Cloud Servers for Lower Carbon. In Proc. of 51st...

    2024

  47. [55]

    Carole-Jean Wu, Ramya Raghavendra, Udit Gupta, Bilge Acun, Newsha Ardalani, Kiwan Maeng, Gloria Chang, Fiona Aga, Jinshi Huang, Charles Bai, et al. 2022. Sustainable AI: Environmental Implications, Challenges and Opportunities. In Proc. of Machine Learning and Systems , Vol. 4...

    2022 doi arXiv

  48. [56]

    Qingyu Xu, Wilson Ricks, Aneesha Manocha, Neha Patankar, and Jesse D Jenkins

  49. [2013]

    In Climate Change 2013: The Physical Science Basis

    Anthropogenic and natural radiative forcing. In Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change . Cambridge University Press, Cambridge, UK, 659–740. https://doi.or...

    2013 doi

  50. [2024]

    Joule 8, 2 (2024), 374–400

    System-level impacts of voluntary carbon-free electricity procurement strategies. Joule 8, 2 (2024), 374–400. 16

    2024

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.