pith. machine review for the scientific record. sign in

arxiv: 2604.08182 · v1 · submitted 2026-04-09 · 💻 cs.DC · cs.AR· cs.ET· cs.PF

Recognition: unknown

Wattlytics: A Web Platform for Co-Optimizing Performance, Energy, and TCO in HPC Clusters

Ayesha Afzal, Georg Hager, Gerhard Wellein

Authors on Pith no claims yet

Pith reviewed 2026-05-10 16:48 UTC · model grok-4.3

classification 💻 cs.DC cs.ARcs.ETcs.PF
keywords GPU performance scalingDVFS power modelingtotal cost of ownershipHPC cluster designenergy efficiencyprocurement decision supportscientific workloads
0
0 comments X

The pith

Energy-efficient GPUs can outperform higher-performance alternatives in cost-effectiveness under budget or energy constraints in HPC clusters.

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

The paper introduces Wattlytics as an interactive browser-based platform that combines benchmark-derived performance scaling for GPUs, power models that vary with frequency settings, and calculations of total cost of ownership over several years. Users can set up different GPU types and workloads to examine outcomes across fixed-budget, fixed-power, fixed-performance, and fixed-count scenarios. Case studies run on the platform indicate that energy-efficient GPUs, when deployed optimally, often produce more work per dollar or per watt than faster but hungrier options. A reader would care because HPC systems consume large amounts of energy and money, so tools that quantify these trade-offs support better procurement choices.

Core claim

Wattlytics integrates benchmark-driven GPU performance scaling, DVFS-aware piecewise power modeling, and multi-year TCO analysis within one interactive environment. It computes metrics including TCO breakdown, work-per-TCO, power-per-TCO, and work-per-watt-per-TCO, while supporting sensitivity analysis through elasticity, Sobol indices, and Monte Carlo methods. Case studies across fixed-budget, fixed-GPU-count, fixed-performance, and fixed-power modes show that optimally deployed energy-efficient GPUs can outperform higher-performance alternatives in overall cost-effectiveness.

What carries the argument

The Wattlytics web platform, which merges benchmark performance data, frequency-sensitive power models, and lifetime cost metrics to enable exploration of heterogeneous GPU cluster designs under constraints.

Load-bearing premise

The benchmark-driven performance scaling and DVFS-aware piecewise power models accurately predict real-world behavior for the selected scientific workloads across the listed GPU architectures under varying frequency settings.

What would settle it

Direct measurements of performance and power draw for workloads such as GROMACS and AMBER on GPUs like H100 and L40S at multiple frequency settings, compared to the platform's model outputs; significant mismatches would show the predictions do not hold.

Figures

Figures reproduced from arXiv: 2604.08182 by Ayesha Afzal, Georg Hager, Gerhard Wellein.

Figure 1
Figure 1. Figure 1: Wattlytics architecture showing the end-to-end pipeline from input to decision analytics. Users provide hardware, benchmark, and cost data, which [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Average application performance and power draw for GROMACS and AMBER at base frequencies and TDP. Each pair of plots shows performance (top) and corresponding power usage (bottom). without exhaustive benchmarking. Single-GPU throughput is frequency-dependent: PGPU(fGPU) = Pbase, GPU · ψGPU(fGPU) = min(P max GPU, b1fGPU + c1). (14) where Pbase, GPU is reference throughput at nominal frequency ( [PITH_FULL_… view at source ↗
Figure 4
Figure 4. Figure 4: TCO breakdown under a fixed ¤10 M budget for GROMACS on Benchmark 4 in Wattlytics; see [S5] experiment link for reproducibility. more L4 or L40S GPUs within the same budget yields 2.5– 4.5× higher lifetime work-per-TCO, confirming that optimal GPU choices are workload-dependent but consistently favor energy-efficient scaling under budget constraints. Upshot 1: Wattlytics exposes non-obvious, multidimension… view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of deployment strategies under different constraints in Wattlytics with a budget cap of [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: (a) What-if electricity price scenario [S11]; (b) Cross-GPU heatmaps comparing elasticity, Sobol indices, and Monte Carlo results, showing the relative [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
read the original abstract

The escalating computational demands and energy footprint of GPU-accelerated computing systems complicate informed design and operational decisions. We present the first release of Wattlytics (https://wattlytics.netlify.app), an interactive, browser-based decision-support system. Unlike existing procurement-oriented calculators, Wattlytics uniquely integrates benchmark-driven GPU performance scaling, dynamic voltage and frequency scaling (DVFS)-aware piecewise power modeling, and multi-year total cost of ownership (TCO) analysis within a single interactive environment. Users can configure heterogeneous systems across contemporary GPU architectures (GH200, H100, L40S, L40, A40, A100, and L4), select representative scientific workloads (e.g., GROMACS, AMBER), and explore deployment scenarios under constraints such as energy prices, system lifetime, and frequency scaling. Wattlytics computes multidimensional decision metrics (TCO breakdown, work-per-TCO, power-per-TCO, and work-per-watt-per-TCO) and supports design-space exploration, what-if scenarios, sensitivity metrics (elasticity, Sobol indices, Monte Carlo) and collaborative features to guide realistic cluster design and procurement under uncertainty. We demonstrate selected scenarios comparing deployment strategies under different operational modes: ixed budget, fixed GPU count, fixed performance, and fixed power. Our case studies show that, under budget or energy constraints, optimally deployed energy-efficient GPUs can outperform higher-performance alternatives in overall cost-effectiveness. Wattlytics helps users explore the design parameter space and distinguish between cost- and risk-driving factors, turning HPC design into a well-informed and explainable decision-making process.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper introduces Wattlytics, a browser-based interactive platform that integrates benchmark-driven GPU performance scaling models, DVFS-aware piecewise power models, and multi-year TCO calculations for heterogeneous HPC clusters. Users can configure systems with GPUs such as GH200, H100, L40S, A100, and L4, select workloads including GROMACS and AMBER, and explore scenarios under fixed-budget, fixed-power, or fixed-performance constraints. The platform outputs multidimensional metrics (TCO breakdown, work-per-TCO, power-per-TCO) plus sensitivity analyses (elasticity, Sobol indices, Monte Carlo). Case studies illustrate that, under budget or energy constraints, optimally deployed energy-efficient GPUs can outperform higher-performance alternatives in cost-effectiveness.

Significance. If the underlying performance scaling and power models are shown to be accurate, Wattlytics offers a genuinely useful, accessible decision-support tool that goes beyond static procurement calculators by enabling what-if exploration, uncertainty quantification, and collaborative analysis. The emphasis on explainable, multi-objective metrics addresses a real gap in HPC design workflows.

major comments (2)
  1. [Case Studies] Case Studies section: the headline finding that energy-efficient GPUs outperform higher-performance alternatives under budget/energy constraints is produced by feeding benchmark-derived scaling curves and frequency-dependent power models into the TCO engine, yet the manuscript supplies no validation against hardware counters, no error bars on predicted performance or power, and no sensitivity of the ranking conclusions to plausible model error. This directly undermines the reliability of the reported multidimensional metrics and deployment recommendations.
  2. [Model description] Model description (performance scaling and DVFS piecewise power models): the abstract and case-study narrative treat these as accurate predictors for the listed workloads across GH200/H100/L40S/etc. at non-default frequencies, but no cross-validation, residual analysis, or comparison to measured data is reported. Because the central claim rests on these models, the absence of quantitative accuracy assessment is load-bearing.
minor comments (2)
  1. [Introduction] The platform URL is given but no permanent archive (e.g., Zenodo snapshot of the deployed code and data) is referenced, which would improve reproducibility of the interactive demonstrations.
  2. [Methods] Notation for the piecewise power model and the exact functional form of the performance scaling curves should be stated explicitly (equations or pseudocode) rather than described only at a high level.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments highlighting the need for explicit model validation and sensitivity analysis. We agree that these elements are important for supporting the reliability of the case-study findings and will revise the manuscript to incorporate them.

read point-by-point responses

  1. Referee: [Case Studies] Case Studies section: the headline finding that energy-efficient GPUs outperform higher-performance alternatives under budget/energy constraints is produced by feeding benchmark-derived scaling curves and frequency-dependent power models into the TCO engine, yet the manuscript supplies no validation against hardware counters, no error bars on predicted performance or power, and no sensitivity of the ranking conclusions to plausible model error. This directly undermines the reliability of the reported multidimensional metrics and deployment recommendations.

    Authors: We acknowledge that the current manuscript does not report hardware-counter validation, error bars, or sensitivity of the GPU rankings to model error. The case studies are presented as demonstrations of the platform's what-if and multi-objective analysis capabilities rather than as definitive procurement advice. In the revision we will add benchmark-derived error bars, a sensitivity study perturbing the scaling and power model parameters within observed variability ranges, and explicit discussion of how these affect the relative cost-effectiveness rankings. This will directly improve the robustness of the reported metrics. revision: yes

  2. Referee: [Model description] Model description (performance scaling and DVFS piecewise power models): the abstract and case-study narrative treat these as accurate predictors for the listed workloads across GH200/H100/L40S/etc. at non-default frequencies, but no cross-validation, residual analysis, or comparison to measured data is reported. Because the central claim rests on these models, the absence of quantitative accuracy assessment is load-bearing.

    Authors: The performance scaling and piecewise power models are constructed from benchmark measurements on the target GPUs, yet the manuscript indeed omits cross-validation, residual analysis, and direct measured-data comparisons. We will add a dedicated subsection (or appendix) that reports these quantitative assessments, including hold-out validation errors, residual distributions, and side-by-side plots of predicted versus measured performance and power for the listed workloads and frequency settings. This will supply the accuracy evaluation needed to underpin the abstract and case-study claims. revision: yes

Circularity Check

0 steps flagged

No circularity: platform integrates empirical models without self-referential reduction

full rationale

The manuscript describes a web-based decision-support tool that combines benchmark-derived performance scaling curves, DVFS-aware piecewise power models, and TCO formulas into interactive metrics. No derivation chain exists in which a claimed prediction or result is shown by the paper's own equations to be identical to its inputs by construction. The case-study outcomes are computed outputs of the integrated models applied to user-configured scenarios; the models themselves are presented as externally benchmarked rather than fitted and then renamed as predictions within the same closed loop. No self-citation load-bearing steps, uniqueness theorems, or ansatzes smuggled via prior work are invoked to justify core claims. The work is therefore self-contained as an engineering platform rather than a mathematical derivation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the accuracy of benchmark-derived performance and power models plus standard TCO parameters (energy prices, lifetime, GPU costs); no explicit free parameters, axioms, or invented entities are identifiable from the abstract alone.

pith-pipeline@v0.9.0 · 5613 in / 1231 out tokens · 53499 ms · 2026-05-10T16:48:58.317420+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

47 extracted references · 26 canonical work pages

  1. [1]

    The Green500 List: Energy-Efficient Supercomputers,

    J. Dongarra, H. Meuer, H. Simon, M. Meuer, and E. Strohmaier, “The Green500 List: Energy-Efficient Supercomputers,” https://www.top500 .org/lists/green500, Nov 2025

    2025

  2. [2]

    A review of energy efficiency evaluation metrics for data centers,

    X. Shao, Z. Zhang, P. Song, Y . Feng, and X. Wang, “A review of energy efficiency evaluation metrics for data centers,”Energy and Buildings, vol. 271, p. 112308, 2022. [Online]. Available: https://doi.org/10.1016/j.enbuild.2022.112308

    work page doi:10.1016/j.enbuild.2022.112308 2022

  3. [3]

    How do data centers make energy efficiency investment decisions? Qualitative evidence from focus groups and interviews,

    H. Klemick, E. Mansur, D. Raimi, and J. Shapiro, “How do data centers make energy efficiency investment decisions? Qualitative evidence from focus groups and interviews,”Energy Efficiency, vol. 12, no. 5, pp. 1359–1377, 2019. [Online]. Available: https: //doi.org/10.1007/s12053-019-09782-2

    work page doi:10.1007/s12053-019-09782-2 2019

  4. [4]

    Accurate Energy and Performance Prediction for Frequency-Scaled GPU Kernels,

    K. Fan, B. Cosenza, and B. Juurlink, “Accurate Energy and Performance Prediction for Frequency-Scaled GPU Kernels,”Computation, vol. 8, no. 2, 2020. [Online]. Available: https://doi.org/10.3390/computation8 020037

    work page doi:10.3390/computation8 2020

  5. [5]

    GPGPU Power Modeling for Multi-domain V oltage-Frequency Scaling,

    J. Guerreiro, A. Ilic, N. Roma, and P. Tomas, “GPGPU Power Modeling for Multi-domain V oltage-Frequency Scaling,” in2018 IEEE International Symposium on High Performance Computer Architecture (HPCA). Los Alamitos, CA, USA: IEEE Computer Society, 2018, pp. 789–800. [Online]. Available: https://doi.org/10.1109/HPCA.2018.0007 2

    work page doi:10.1109/hpca.2018.0007 2018

  6. [6]

    A survey and measurement study of GPU DVFS on energy conservation,

    X. Mei, Q. Wang, and X. Chu, “A survey and measurement study of GPU DVFS on energy conservation,”Digital Communications and Networks, vol. 3, no. 2, pp. 89–100, 2017. [Online]. Available: https://doi.org/10.1016/j.dcan.2016.10.001

    work page doi:10.1016/j.dcan.2016.10.001 2017

  7. [7]

    Rogers, Tor M

    V . Kandiah, S. Peverelle, M. Khairy, J. Pan, A. Manjunath, T. G. Rogers, T. M. Aamodt, and N. Hardavellas, “AccelWattch: A Power Modeling Framework for Modern GPUs,” inMICRO-54: 54th Annual IEEE/ACM International Symposium on Microarchitecture. New York, NY , USA: Association for Computing Machinery, 2021, p. 738–753. [Online]. Available: https://doi.org...

    work page doi:10.1145/3466752.3480063 2021

  8. [8]

    PowerSensor3: A Fast and Accurate Open Source Power Measurement Tool,

    S. van der Vlugt, L. Oostrum, G. Schoonderbeek, B. van Werkhoven, B. Veenboer, K. Doekemeijer, and J. Romein, “PowerSensor3: A Fast and Accurate Open Source Power Measurement Tool,” inProceedings of the International Symposium on Performance Analysis of Systems and Software (ISPASS), 2025. [Online]. Available: https://doi.org/10.48550/arXiv.2504.17883

    work page doi:10.48550/arxiv.2504.17883 2025

  9. [9]

    EAR: Energy management framework for HPC,

    J. Corbal ´an and L. Brochard, “EAR: Energy management framework for HPC,” https://www.bsc.es/research-and-development/software-and -apps/software-list/ear-energy-management-framework-hpc, 2018

    2018

  10. [10]

    Lee, Meng Li, Bert Maher, Dheevatsa Mudigere, Maxim Naumov, Martin Schatz, Mikhail Smelyanskiy, Xiaodong Wang, Brandon Reagen, Carole-Jean Wu, Mark Hemp- stead, and Xuan Zhang

    M. Khairy, Z. Shen, T. M. Aamodt, and T. G. Rogers, “Accel-Sim: An Extensible Simulation Framework for Validated GPU Modeling,” in2020 ACM/IEEE 47th Annual International Symposium on Computer Architecture (ISCA), 2020, pp. 473–486. [Online]. Available: https://doi.org/10.1109/ISCA45697.2020.00047

    work page doi:10.1109/isca45697.2020.00047 2020

  11. [11]

    Powerlog: Lightweight Power Profiling Tool for NVIDIA GPUs,

    A. R. Shovon, “Powerlog: Lightweight Power Profiling Tool for NVIDIA GPUs,” https://pypi.org/project/powerlog, 2026

    2026

  12. [12]

    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, 2010, pp. 207–216. [Online]. Available: https://doi.org/10.1109/ICPP W.2010.38

    work page doi:10.1109/icpp 2010

  13. [13]

    AIMeter: Measuring, Analyzing, and Visualizing Energy and Carbon Footprint of AI Workloads,

    H. Huang, K. Zhang, H. Liao, K. Wu, and G. Tang, “AIMeter: Measuring, Analyzing, and Visualizing Energy and Carbon Footprint of AI Workloads,” inArXiv preprint, 2025. [Online]. Available: https://doi.org/10.48550/arXiv.2506.20535

    work page doi:10.48550/arxiv.2506.20535 2025

  14. [14]

    WattScope: Non-intrusive Application-level Power Disaggregation in Datacenters,

    X. Guan, N. Bashir, D. Irwin, and P. Shenoy, “WattScope: Non-intrusive Application-level Power Disaggregation in Datacenters,”SIGMETRICS Perform. Eval. Rev., vol. 51, no. 4, p. 24–25, Feb. 2024. [Online]. Available: https://doi.org/10.1145/3649477.3649491

    work page doi:10.1145/3649477.3649491 2024

  15. [15]

    mlco2/codecarbon: v3.2.6,

    B. Courty, V . Schmidt, S. Luccioni, Goyal-Kamal, MarionCoutarel, B. Feld, J. Lecourt, LiamConnell, A. Saboni, Inimaz, supatomic, M. L ´eval, L. Blanche, A. Cruveiller, ouminasara, F. Zhao, A. Joshi, A. Bogroff, H. de Lavoreille, N. Laskaris, E. Abati, D. Blank, Z. Wang, A. Catovic, M. Alencon, M. Stechly, C. Bauer, L. O. N. de Ara ´ujo, JPW, and MinervaB...

  16. [16]

    Available: https://doi.org/10.5281/zenodo.19334697

    [Online]. Available: https://doi.org/10.5281/zenodo.19334697

    work page doi:10.5281/zenodo.19334697

  17. [17]

    A Simple Model for Determining True Total Cost of Ownership for Data Centers,

    J. Koomey, K. Brill, P. Turner, J. Stanley, and B. Taylor, “A Simple Model for Determining True Total Cost of Ownership for Data Centers,” https://m.softchoice.com/files/pdf/about/sustain-enable/simplemodeldet ermingtruetco.pdf, 2007

    2007

  18. [18]

    Total Cost of Ownership (TCO) resources and calculators,

    NVIDIA, “Total Cost of Ownership (TCO) resources and calculators,” https://www.nvidia.com/en-us/networking/total-cost-ownership, 2026

    2026

  19. [19]

    Intel Xeon Processor Advisor,

    Intel, “Intel Xeon Processor Advisor,” https://xeonprocessoradvisor.int el.com, 2026

    2026

  20. [20]

    AMD EPYC Server Virtualization TCO Estimation Tool,

    AMD, “AMD EPYC Server Virtualization TCO Estimation Tool,” https: //www.amd.com/en/processors/epyc-VirtTCOtool, 2026

    2026

  21. [21]

    Total Cost of Ownership (TCO) Calculator,

    Scale Computing, “Total Cost of Ownership (TCO) Calculator,” https: //www.scalecomputing.com/total-cost-of-ownership-tco-calculator, 2026

    2026

  22. [22]

    Cloud Carbon Footprint,

    ThoughtWorks and the Cloud Carbon Footprint community, “Cloud Carbon Footprint,” https://www.cloudcarbonfootprint.org/ and https://github.com/cloud-carbon-footprint/cloud-carbon-footprint, 2026

    2026

  23. [23]

    DC Pro: Data Center Profiler,

    Lawrence Berkeley National Laboratory (LBNL), “DC Pro: Data Center Profiler,” https://datacenters.lbl.gov/dcpro, 2026

    2026

  24. [24]

    LT-TCO: A TCO Calculation Model of Data Centers for Long-Term Data Preservation,

    W. Yan, J. Yao, Q. Cao, and Y . Zhang, “LT-TCO: A TCO Calculation Model of Data Centers for Long-Term Data Preservation,” in2019 IEEE International Conference on Networking, Architecture and Storage (NAS), 2019, pp. 1–8. [Online]. Available: https: //doi.org/10.1109/NAS.2019.8834714

    work page doi:10.1109/nas.2019.8834714 2019

  25. [25]

    Metrics and an Infrastructure Model to Evaluate Data Center Efficiency,

    C. L. Belady and C. G. Malone, “Metrics and an Infrastructure Model to Evaluate Data Center Efficiency,” inProceedings of the ASME 2007 InterPACK Conference collocated with the ASME/JSME 2007 Thermal Engineering Heat Transfer Summer Conference, ser. International Electronic Packaging Technical Conference and Exhibition, vol. 1, 2007, pp. 751–755. [Online]...

    work page doi:10.1115/ipack2007-33338 2007

  26. [26]

    Algorithm for Calculating TCO and SCE Metrics to Assess the Efficiency of Using a Data Center,

    B. Denisenko, M. Tyanutov, I. Nikiforov, and S. Ustinov, “Algorithm for Calculating TCO and SCE Metrics to Assess the Efficiency of Using a Data Center,” in2nd International Conference on Computer Applications for Management and Sustainable Development of Production and Industry (CMSD-II-2022), S. Sadullozoda and A. Gibadullin, Eds., vol. 12564, Internati...

    work page doi:10.1117/12.2669285 2022

  27. [27]

    SPEC Power Benchmark,

    Standard Performance Evaluation Corporation (SPEC), “SPEC Power Benchmark,” https://www.spec.org/power ssj2008, 2026

    2026

  28. [28]

    SPEC SERT: Server Efficiency Rating Tool,

    ——, “SPEC SERT: Server Efficiency Rating Tool,” https://www.spec .org/sert, 2026

    2026

  29. [29]

    MLPerf Power Benchmark,

    MLPerf, “MLPerf Power Benchmark,” https://mlperf.org/power, 2026

    2026

  30. [30]

    HPCTOOLKIT: tools for performance analysis of optimized parallel programs,

    L. Adhianto, S. Banerjee, M. Fagan, M. Krentel, G. Marin, J. Mellor-Crummey, and N. R. Tallent, “HPCTOOLKIT: tools for performance analysis of optimized parallel programs,”Concurrency and Computation: Practice and Experience, vol. 22, no. 6, pp. 685–701,

  31. [31]

    Available: https://www.doi.org/10.1002/cpe.1553

    [Online]. Available: https://doi.org/10.1002/cpe.1553

    work page doi:10.1002/cpe.1553

  32. [32]

    The cost of computation: Metrics and models for modern multicore-based systems in scientific computing,

    A. Afzal, “The cost of computation: Metrics and models for modern multicore-based systems in scientific computing,”Master’s thesis, Department Informatik, Friedrich Alexander Universit ¨at Erlangen- N¨urnberg, 2015. [Online]. Available: https://doi.org/10.13140/RG.2.2. 35954.25283

    work page doi:10.13140/rg.2.2 2015

  33. [33]

    SPEChpc 2021 Benchmarks on Ice Lake and Sapphire Rapids Infiniband Clusters: A Performance and Energy Case Study,

    A. Afzal, G. Hager, and G. Wellein, “SPEChpc 2021 Benchmarks on Ice Lake and Sapphire Rapids Infiniband Clusters: A Performance and Energy Case Study,” in14th IEEE/ACM Performance Modeling, Benchmarking and Simulation of High Performance Computer Systems (PMBS), 2023. [Online]. Available: https://doi.org/10.1145/3624062.36 24197

    work page doi:10.1145/3624062.36 2021

  34. [34]

    Analytic Roofline Modeling and Energy Analysis of LULESH Proxy Application on Multi-Core Clusters,

    ——, “Analytic Roofline Modeling and Energy Analysis of LULESH Proxy Application on Multi-Core Clusters,”International Journal of High Performance Computing Applications (IJHPCA), 2025. [Online]. Available: https://doi.org/10.1177/10943420251363711

    work page doi:10.1177/10943420251363711 2025

  35. [35]

    NVIDIA GPU Architecture Specifications and Datasheets,

    NVIDIA Corporation, “NVIDIA GPU Architecture Specifications and Datasheets,” https://www.nvidia.com/en-us/data-center, 2026

    2026

  36. [36]

    AWS Pricing Calculator,

    Amazon Web Services (AWS), “AWS Pricing Calculator,” https://calc ulator.aws, 2026

    2026

  37. [37]

    Google Cloud Pricing Calculator,

    Google Cloud, “Google Cloud Pricing Calculator,” https://cloud.google .com/products/calculator, 2026

    2026

  38. [38]

    Azure Pricing Calculator,

    Microsoft Azure, “Azure Pricing Calculator,” https://azure.microsoft.co m/en-us/pricing/calculator, 2026

    2026

  39. [39]

    ENERGY STAR Program for Data Center Equip- ment,

    ENERGY STAR, “ENERGY STAR Program for Data Center Equip- ment,” https://www.energystar.gov/products/data center equipment, 2026

    2026

  40. [40]

    Life Cycle Analysis for Emissions of Scientific Computing Centres,

    M. Wadenstein and W. Vanderbauwhede, “Life Cycle Analysis for Emissions of Scientific Computing Centres,”The European Physical Journal C, vol. 85, p. 913, 2025. [Online]. Available: https://doi.org/10.1140/epjc/s10052-025-14650-8

    work page doi:10.1140/epjc/s10052-025-14650-8 2025

  41. [41]

    Navigating Energy Doldrums: Modeling the Impact of Energy Price V olatility on HPC Cost of Ownership,

    P. Arzt and F. Wolf, “Navigating Energy Doldrums: Modeling the Impact of Energy Price V olatility on HPC Cost of Ownership,”ArXiv preprint, 2025. [Online]. Available: https://arxiv.org/abs/2509.07567

    work page arXiv 2025

  42. [42]

    GROMACS Unplugged: How Power Capping and Frequency Shapes Performance on GPUs,

    A. Afzal, A. Kahler, G. Hager, and G. Wellein, “GROMACS Unplugged: How Power Capping and Frequency Shapes Performance on GPUs,”Euro-Par 2025: Parallel Processing Workshops Volume in the Springer Lecture Notes in Computer Science (LNCS) series, 2025. [Online]. Available: https://doi.org/10.48550/arXiv.2412.08792

    work page doi:10.48550/arxiv.2412.08792 2025

  43. [43]

    An Integrated GPU Power and Performance Model,

    S. Hong and H. Kim, “An Integrated GPU Power and Performance Model,” inProceedings of the 37th Annual International Symposium on Computer Architecture (ISCA), 2010, pp. 280–289. [Online]. Available: https://doi.org/10.1145/1815961.1815998

    work page doi:10.1145/1815961.1815998 2010

  44. [44]

    Experience on Clock Rate Adjustment for Energy-Efficient GPU-Accelerated Real-World Codes,

    G. Amati, M. Turisini, A. Monterubbiano, M. Paladino, E. Boella, D. Gregori, and D. Croce, “Experience on Clock Rate Adjustment for Energy-Efficient GPU-Accelerated Real-World Codes,” inHigh Performance Computing. Cham: Springer Nature Switzerland, 2026, pp. 245–257. [Online]. Available: https://doi.org/10.1007/978-3-032-0 7612-0 19

    work page doi:10.1007/978-3-032-0 2026

  45. [45]

    Global sensitivity indices for nonlinear mathematical models and their Monte Carlo estimates,

    I. Sobol, “Global sensitivity indices for nonlinear mathematical models and their Monte Carlo estimates,”Mathematics and Computers in Simulation, vol. 55, no. 1, pp. 271–280, 2001. [Online]. Available: https://doi.org/10.1016/S0378-4754(00)00270-6

    work page doi:10.1016/s0378-4754(00)00270-6 2001

  46. [46]

    Variance based sensitivity analysis of model output. Design and estimator for the total sensitivity index,

    A. Saltelli, P. Annoni, I. Azzini, F. Campolongo, M. Ratto, and S. Tarantola, “Variance based sensitivity analysis of model output. Design and estimator for the total sensitivity index,”Computer Physics Communications, vol. 181, no. 2, pp. 259–270, 2010. [Online]. Available: https://doi.org/10.1016/j.cpc.2009.09.018

    work page doi:10.1016/j.cpc.2009.09.018 2010

  47. [47]

    Slurm Energy Plugin,

    “Slurm Energy Plugin,” https://slurm.schedmd.com/slurm.conf.html, 2026. SITOGRAPHY VII. WATTLYTICSREPRODUCIBILITYSITOGRAPHY [S1] https://tinyurl.com/Wattlytics-R1 [S2] https://tinyurl.com/Wattlytics-R2 [S3] https://tinyurl.com/Wattlytics-R3 [S4] https://tinyurl.com/Wattlytics-R4 [S5] https://tinyurl.com/Wattlytics-R5 [S6] https://tinyurl.com/Wattlytics-R6...

    2026