Pith. sign in

REVIEW 2 major objections 5 minor 35 references

Choosing the right number of threads can cut the energy cost of HEP Monte Carlo events by tens of percent with only a few-percent runtime penalty.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-11 10:01 UTC pith:TLIF6T7I

load-bearing objection Solid engineering paper: watts-per-event metric + Docker toolbox + multi-CPU data show energy-optimal thread counts often beat pure speed for HIJING++, with only a few-percent runtime hit. the 2 major comments →

arxiv 2607.05018 v1 pith:TLIF6T7I submitted 2026-07-06 physics.comp-ph cs.DChep-ph

Watts per event: evaluating Sustainability of HEP Event Generators beyond the LHC era

Szabolcs Moln\'ar , G\'abor B\'ir\'o , G\'abor Papp , Gergely G\'abor Barnaf\"oldi This is my paper
classification physics.comp-ph cs.DChep-ph
keywords Monte Carlo event generatorsenergy efficiencymultithreadingHIJING++HEP computingsustainabilityRAPL power measurementHL-LHC
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Monte Carlo event generators for high-energy physics are among the largest consumers of CPU time on worldwide computing grids, and that demand is expected to rise by an order of magnitude for the High-Luminosity LHC and Future Circular Collider. The paper shows that the energy spent to produce one physics event is not automatically minimized by using every available core. Instead, a simple cost metric—average package power divided by event throughput—reveals a clear sweet spot in the number of threads for each CPU architecture and collision system. On modern processors the energy-optimal configuration can save roughly 30 percent of the power relative to single-thread runs while lengthening wall-clock time by only a few percent compared with the absolute-fastest thread count. A ready-to-run Docker toolbox is supplied so that any group can measure the same quantity on its own hardware and generators. The practical payoff is that large-scale tuning campaigns, which already require tens of billions of events, can be made substantially greener without rewriting the underlying physics code.

Core claim

The energy cost per generated event, defined as average CPU package power divided by event throughput, is a non-monotonic function of the number of threads; for every architecture and drive type examined there exists a thread count that minimizes this cost, and that optimum is often lower than the thread count that minimizes wall-clock time. On the newest EPYC processor the energy saving relative to single-thread Pb-Pb production reaches 29–34 percent while the runtime penalty versus the absolute-fastest configuration stays under 5 percent for most storage setups.

What carries the argument

The per-event energy price E_event = P_avg / T_event, measured under exclusive use of the CPU package via RAPL counters inside a privileged container, is the central figure of merit that turns raw power and throughput curves into a concrete optimisation target.

Load-bearing premise

That the RAPL package-power reading taken under exclusive Docker control, while ignoring memory and disk power, is a faithful enough proxy for the true energy cost of production campaigns on shared grid resources.

What would settle it

Re-run the same HIJING++ benchmarks on a production WLCG node that also records whole-node power (including DRAM and storage) and check whether the thread count that minimises package-only E_event still minimises total energy per event.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Tuning campaigns that already generate O(10^10) events can cut their electricity bill by tens of percent simply by locking the thread count to the measured optimum rather than to the core count.
  • The same E_event scan can be applied to any other parallel Monte Carlo generator (Pythia, Herwig, Sherpa, etc.) without modifying the physics code.
  • Hardware procurement for future HL-LHC and FCC computing can be scored by energy-per-event at the optimal thread count rather than by peak FLOPS or TDP alone.
  • Drive type (SSD vs HDD) has only a secondary effect on the energy optimum for modern CPUs, so storage upgrades are less urgent than thread-count tuning.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same non-monotonic energy-versus-thread curve is likely to appear in any memory-bandwidth-sensitive HEP workflow (reconstruction, pile-up overlay, lattice QCD), so the method is not limited to generators.
  • If the RAPL-only approximation systematically underestimates total power, the absolute energy numbers will shift but the location of the optimum may remain stable; a cheap whole-node power meter would settle the question.
  • Publishing the optimal-thread table for each new CPU generation would give sites an immediate, zero-code-change lever for carbon-aware scheduling.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The manuscript introduces a practical energy-efficiency metric for Monte Carlo event generators, E_event = P_avg / T_event (Eq. 1), and applies it to the multithreaded HIJING++ heavy-ion generator using a containerized toolbox (77rev/proripy). Direct RAPL-based power and throughput measurements are reported for four CPU architectures (Xeon E5-2650, EPYC 7502P, Ryzen 7 8845HS, EPYC 4585PX), three I/O modes, and both HL-LHC and FCC energies in pp and Pb-Pb systems. The central empirical result is that the thread count minimizing energy per event frequently differs from the absolute-fastest configuration, yielding energy savings of order 30 % relative to single-thread (and a few-percent runtime penalty versus the fastest multi-thread setting) on modern CPUs (Tables 3, 4, 7, 8). Illustrative HIJING++ tuning results and campaign-scale energy estimates are also provided.

Significance. If the reported optima hold under production conditions, the work supplies a concrete, immediately usable method for reducing the energy footprint of one of the most CPU-intensive components of the HEP computing pipeline. The public Docker image, the explicit benchmarking scripts, and the tabulated optimal-versus-fastest comparisons constitute reproducible engineering results that can be applied to other generators and hardware. The demonstration that energy-optimal thread counts can differ from runtime-optimal ones on recent architectures is a useful practical finding for HL-LHC and FCC-scale campaigns.

major comments (2)
  1. The energy estimates for full tuning campaigns (Tables 5 and 6) are simple multiplications of the measured single-run rates by a fixed number of configurations, events and iterations. While the arithmetic is transparent, the manuscript does not quantify how sensitive those totals are to the (unmeasured) overhead of the full Professor–Rivet–YODA pipeline, to thermal throttling under sustained multi-day loads, or to the neglected RAM/storage power already noted in §4. A short sensitivity discussion or an explicit statement that the tables are order-of-magnitude illustrations would strengthen the claim that the optimal-thread choice yields ‘substantial’ long-term savings.
  2. All power data are obtained under a privileged Docker container with no other user processes (§4). The relative ranking of thread counts is therefore clean, but the absolute E_event values and the claimed transferability to shared WLCG nodes remain untested. A single cross-check on bare metal or under a realistic multi-user load would make the sustainability argument more robust; without it the absolute numbers should be presented more cautiously.
minor comments (5)
  1. The Docker image is named 77rev/proripy in the body and 77rev/propripy in the abstract; the two spellings should be unified.
  2. Figures 1–4 are described as ‘compact’ and limited to 10 threads for visibility, yet the text and Table 3 discuss optima up to 15 threads. Either extend the plotted range or add a clear note that the full data appear only in the appendix figures.
  3. Table 2 lists TDP values while the text carefully distinguishes TDP from the RAPL package power actually measured; a one-sentence reminder in the table caption would avoid confusion.
  4. A few typographical inconsistencies remain (e.g., ‘HL-HLC’ in a figure caption, ‘Xenon’ for Xeon in an appendix figure label).
  5. The initialization-time data (Table A1, Fig. A1) are useful but never referenced in the main discussion of energy cost; a brief remark on whether they are included in the E_event averages would clarify the accounting.

Circularity Check

0 steps flagged

No significant circularity: E_event is an empirical ratio of independently measured power and throughput; optima are read off the resulting curves.

full rationale

The paper’s central claim is an engineering measurement, not a derivation. Equation (1) defines E_event = P_avg / T_event from two quantities obtained by separate instrumentation (RAPL package power via powerstat and event throughput under a dedicated Docker run). Optimal thread counts are simply the minima of the measured curves (Figs. 2, 4; Tables 3, 4). No parameter is fitted to data and then re-used as a “prediction,” no uniqueness theorem is imported from prior self-citations, and the HIJING++ tuning results of §6 are explicitly labelled illustrative. Self-citations to earlier HIJING++ papers merely identify the generator under test; they do not underwrite the energy-efficiency metric. The derivation chain is therefore self-contained against external benchmarks and contains no circular reduction.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 1 invented entities

The central claim rests on a small set of standard measurement assumptions and one operational definition; no free parameters are fitted to produce the energy-optimum result, and no new physical entities are postulated.

axioms (4)
  • domain assumption RAPL package energy counters (read by powerstat) accurately report the electrical power drawn by the CPU package under the workload.
    Stated in §4; the entire E_event metric is built on these readings.
  • domain assumption Power drawn by RAM, storage and other system components can be neglected relative to CPU package power for the purpose of ranking thread configurations.
    Explicitly justified in §4 as a simplification; if false, absolute energy numbers shift but relative rankings may still hold.
  • domain assumption When the CPU is dedicated to the generator (no other user processes), measured throughput and power are representative of production campaign efficiency.
    Stated in §2 and §4; shared-cluster contention could alter the location of the energy optimum.
  • ad hoc to paper E_event ≡ P_avg / T_event is a valid scalar figure of merit for sustainability of event generation.
    Defined in Eq. (1); the paper’s optimization claim is with respect to this quantity.
invented entities (1)
  • 77rev/proripy Docker toolbox independent evidence
    purpose: Provides a reproducible, version-pinned environment containing Rivet, Professor, Pythia, ROOT, HepMC3 and the power-measurement scripts used for all benchmarks.
    Introduced in §4 and Table 1; it is an engineering artifact, not a physical postulate, and is independently downloadable.

pith-pipeline@v1.1.0-grok45 · 16372 in / 2740 out tokens · 26918 ms · 2026-07-11T10:01:30.543936+00:00 · methodology

0 comments
read the original abstract

The development, tuning and operation of Monte Carlo event generators beyond the LHC era require vast amount of resources. In this study we investigate the sustainability of these software with a containerized set of tools (named 77rev/propripy), by benchmarking the HIJING++ heavy-ion Monte Carlo event generator. We analyze the performance of various CPU architectures and show that by choosing the level of multithreading properly, the cost of event generation can be optimized. The presented approach can reduce the energy footprint of high-energy physics event generators and therefore alleviate the ever-increasing, ubiquitous computational challenges.

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

35 extracted references · 4 canonical work pages · 3 internal anchors

  1. [1]

    Zurbano Fernandez, I., et al.: High-Luminosity Large Hadron Collider (HL- LHC): Technical design report10/2020(2020) https://doi.org/10.23731/ CYRM-2020-0010

    2020

  2. [2]

    Agostini, P.,et al.: The Large Hadron–Electron Collider at the HL-LHC. J. Phys. G48(11), 110501 (2021) https://doi.org/10.1088/1361-6471/abf3ba arXiv:2007.14491 [hep-ex]

    doi:10.1088/1361-6471/abf3ba 2021

  3. [3]

    Abada, A.,et al.: FCC-ee: The Lepton Collider: Future Circular Collider Con- ceptual Design Report Volume 2. Eur. Phys. J. ST228(2), 261–623 (2019) https://doi.org/10.1140/epjst/e2019-900045-4

    doi:10.1140/epjst/e2019-900045-4 2019

  4. [4]

    Abada, A.,et al.: FCC Physics Opportunities: Future Circular Collider Con- ceptual Design Report Volume 1. Eur. Phys. J. C79(6), 474 (2019) https: //doi.org/10.1140/epjc/s10052-019-6904-3

    doi:10.1140/epjc/s10052-019-6904-3 2019

  5. [5]

    (eds.): LHC computing Grid

    Bird, I., et al. (eds.): LHC computing Grid. Technical design report (2005)

    2005

  6. [6]

    Aad, G.,et al.: The environmental impact, carbon emissions and sustainability of computing in the ATLAS experiment. Eur. Phys. J. C85(12), 1397 (2025) https://doi.org/10.1140/epjc/s10052-025-14976-3 arXiv:2505.08530 [hep-ex]

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1140/epjc/s10052-025-14976-3 2025

  7. [7]

    Acharya, S.,et al.: The ALICE experiment: a journey through QCD. Eur. Phys. J. C84(8), 813 (2024) https://doi.org/10.1140/epjc/s10052-024-12935-y arXiv:2211.04384 [nucl-ex] 14

    Pith/arXiv arXiv doi:10.1140/epjc/s10052-024-12935-y 2024

  8. [8]

    Buckley, A., Hoeth, H., Lacker, H., Schulz, H., Seggern, J.E.: Systematic event generator tuning for the LHC. Eur. Phys. J. C65, 331–357 (2010) https://doi. org/10.1140/epjc/s10052-009-1196-7 arXiv:0907.2973 [hep-ph]

    Pith/arXiv arXiv doi:10.1140/epjc/s10052-009-1196-7 2010

  9. [9]

    Lazzarin, M., Alioli, S., Carrazza, S.: MCNNTUNES: Tuning Shower Monte Carlo generators with machine learning. Comput. Phys. Commun.263, 107908 (2021) https://doi.org/10.1016/j.cpc.2021.107908 arXiv:2010.02213 [physics.comp-ph]

    doi:10.1016/j.cpc.2021.107908 2021

  10. [10]

    JINST3, 08001 (2008) https://doi.org/10

    Evans, L., Bryant, P.: LHC Machine. JINST3, 08001 (2008) https://doi.org/10. 1088/1748-0221/3/08/S08001

    2008

  11. [11]

    Szczepanek, N., Britton, D., Di Girolamo, A., Ketele, E., Glushkov, I., Gior- dano, D., Ondris, L., Simili, E., Borge, G.M.: HEP Benchmark Suite: Enhancing Efficiency and Sustainability in Worldwide LHC Computing Infrastructures. (2024)

    2024

  12. [12]

    SciPost Phys

    Bierlich, C., Buckley, A., Butterworth, J.M., Gutschow, C., Lonnblad, L., Procter, T., Richardson, P., Yeh, Y.: Robust independent validation of experiment and theory: Rivet version 4 release note. SciPost Phys. Codeb.36, 1 (2024) https: //doi.org/10.21468/SciPostPhysCodeb.36 arXiv:2404.15984 [hep-ph]

    Pith/arXiv arXiv doi:10.21468/scipostphyscodeb.36 2024

  13. [13]

    Docker Hub

    Moln´ ar, S., B´ ır´ o, G., Barnaf¨ oldi, G.G.: 77rev/proripy Docker Image. Docker Hub. Tag: 4.4, [Online; accessed 01. 07. 2026] (2026). https://hub.docker.com/r/77rev/ proripy

    2026

  14. [15]

    Albacete, J.L.,et al.: Predictions for Cold Nuclear Matter Effects inp+Pb Colli- sions at √sN N = 8.16 TeV. Nucl. Phys.A972, 18–85 (2018) https://doi.org/10. 1016/j.nuclphysa.2017.11.015 arXiv:1707.09973 [hep-ph]

    Pith/arXiv arXiv 2018

  15. [16]

    Barnaf¨ oldi, G.G., B´ ır´ o, G., Gyulassy, M., Harangoz´ o, S.M., L´ evai, P., Ma, G., Papp, G., Wang, X.-N., Zhang, B.-W.: First Results with HIJING++ in High- Energy Heavy-Ion Collisions. Nucl. Part. Phys. Proc.289-290, 373–376 (2017) https://doi.org/10.1016/j.nuclphysbps.2017.05.086 arXiv:1701.08496 [hep-ph]

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1016/j.nuclphysbps.2017.05.086 2017

  16. [17]

    PoS High-pT (015) (2018) arXiv:1805.02635 [hep-ph]

    Papp, G., Barnaf¨ oldi, G.G., B´ ır´ o, G., Gyulassy, M., Harangoz´ o, S., Ma, G., L´ evai, P., Wang, X.-N., Zhang, B.-W.: First Results with HIJING++ on High-energy Heavy Ion Collisions. PoS High-pT (015) (2018) arXiv:1805.02635 [hep-ph]

    Pith/arXiv arXiv 2018

  17. [18]

    MDPI Proc.10(1), 4 (2019) arXiv:1811.02131 [physics.comp-ph] 15

    B´ ır´ o, G., Papp, G., Barnaf¨ oldi, G.G., Nagy, D., Gyulassy, M., L´ evai, P., Wang, X.-N., Zhang, B.-W.: HIJING, a Heavy Ion Jet INteraction Generator for the High-Luminosity Era of the LHC and Beyond. MDPI Proc.10(1), 4 (2019) arXiv:1811.02131 [physics.comp-ph] 15

    Pith/arXiv arXiv 2019

  18. [19]

    Universe5(6), 134 (2019) arXiv:1905.11272 [hep-ph]

    B´ ır´ o, G., G´ abor Barnaf¨ oldi, G., Papp, G., S´ andor Bir´ o, T.: Multiplicity Depen- dence in the Non-Extensive Hadronization Model Calculated by the HIJING Framework++. Universe5(6), 134 (2019) arXiv:1905.11272 [hep-ph]

    Pith/arXiv arXiv 2019

  19. [20]

    Introducing HIJING++: the Heavy Ion Monte Carlo Generator for the High-Luminosity LHC Era

    B´ ır´ o, G., Barnaf¨ oldi, G.G., Papp, G., Gyulassy, M., L´ evai, P., Wang, X.-N., Zhang, B.-W.: Introducing HIJING++: the Heavy Ion Monte Carlo Generator for the High-Luminosity LHC Era. PoSHardProbes2018, 045 (2019) https://doi.org/ 10.22323/1.345.0045 arXiv:1901.04220 [physics.comp-ph]

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.22323/1.345.0045 2019

  20. [21]

    Sjostrand, T.: The Lund Monte Carlo for Jet Fragmentation. Comput. Phys. Commun.27, 243 (1982) https://doi.org/10.1016/0010-4655(82)90175-8

    work page doi:10.1016/0010-4655(82)90175-8 1982

  21. [22]

    Sj¨ ostrand, T., Ask, S., Christiansen, J.R., Corke, R., Desai, N., Ilten, P., Mrenna, S., Prestel, S., Rasmussen, C.O., Skands, P.Z.: An introduction to PYTHIA 8.2. Comput. Phys. Commun.191, 159–177 (2015) https://doi.org/10.1016/j.cpc. 2015.01.024 arXiv:1410.3012 [hep-ph]

    Pith/arXiv arXiv doi:10.1016/j.cpc 2015

  22. [23]

    Skands, P., Carrazza, S., Rojo, J.: Tuning PYTHIA 8.1: the Monash 2013 Tune. Eur. Phys. J. C74(8), 3024 (2014) https://doi.org/10.1140/epjc/ s10052-014-3024-y arXiv:1404.5630 [hep-ph]

    Pith/arXiv arXiv doi:10.1140/epjc/ 2013

  23. [24]

    JHEP 08, 003 (2015) https://doi.org/10.1007/JHEP08(2015)003 arXiv:1505.01681 [hep-ph]

    Christiansen, J.R., Skands, P.Z.: String Formation Beyond Leading Colour. JHEP 08, 003 (2015) https://doi.org/10.1007/JHEP08(2015)003 arXiv:1505.01681 [hep-ph]

    Pith/arXiv arXiv doi:10.1007/jhep08(2015)003 2015

  24. [25]

    Sj¨ ostrand, T.: The Development of MPI Modeling in Pythia. Adv. Ser. Direct. High Energy Phys.29, 191–225 (2018) https://doi.org/10.1142/9789813227767 0010 arXiv:1706.02166 [hep-ph]

    Pith/arXiv arXiv doi:10.1142/9789813227767 2018

  25. [26]

    JHEP10, 134 (2018) https://doi.org/10

    Bierlich, C., Gustafson, G., L¨ onnblad, L., Shah, H.: The Angantyr model for Heavy-Ion Collisions in PYTHIA8. JHEP10, 134 (2018) https://doi.org/10. 1007/JHEP10(2018)134 arXiv:1806.10820 [hep-ph]

    Pith/arXiv arXiv 2018

  26. [27]

    SciPost Phys

    Bierlich, C.,et al.: A comprehensive guide to the physics and usage of PYTHIA 8.3. SciPost Phys. Codeb.2022, 8 (2022) https://doi.org/10.21468/ SciPostPhysCodeb.8 arXiv:2203.11601 [hep-ph]

    Pith/arXiv arXiv 2022

  27. [28]

    [Online; accessed 01

    Toolbox for HEP. [Online; accessed 01. 07. 2026]. https://gitlab.wigner.hu/ szmlnr/toolbox-for-hep

    2026

  28. [29]

    https://www.techpowerup.com/cpu-specs/xeon-e5-2650

    Intel Xeon E5-2650. https://www.techpowerup.com/cpu-specs/xeon-e5-2650. c974. [Online; accessed 01. 07. 2026]

    2026

  29. [30]

    https://www.techpowerup.com/cpu-specs/epyc-7502p

    AMD EPYC 7502P. https://www.techpowerup.com/cpu-specs/epyc-7502p. c2260. [Online; accessed 01. 07. 2026]

    2026

  30. [31]

    https://www.techpowerup.com/cpu-specs/ 16 ryzen-7-8845hs.c3400

    AMD Ryzen 7 8845HS. https://www.techpowerup.com/cpu-specs/ 16 ryzen-7-8845hs.c3400. [Online; accessed 01. 07. 2026]

    2026

  31. [32]

    https://www.techpowerup.com/cpu-specs/epyc-4585px

    AMD EPYC 4585PX. https://www.techpowerup.com/cpu-specs/epyc-4585px. c4172. [Online; accessed 01. 07. 2026]

    2026

  32. [33]

    Alver, B.,et al.: Charged-particle multiplicity and pseudorapidity distributions measured with the phobos detector in Au + Au, Cu + Cu,d+ Au, andp+p collisions at ultrarelativistic energies. Phys. Rev. C83, 024913 (2011) https:// doi.org/10.1103/PhysRevC.83.024913

    doi:10.1103/physrevc.83.024913 2011

  33. [34]

    Adam, J.,et al.: Charged-particle multiplicities in proton–proton collisions at√s= 0.9 to 8 TeV. Eur. Phys. J. C77(1), 33 (2017) https://doi.org/10.1140/ epjc/s10052-016-4571-1 arXiv:1509.07541 [nucl-ex]

    Pith/arXiv arXiv 2017

  34. [35]

    JHEP11, 013 (2018) https://doi.org/10.1007/JHEP11(2018)013 arXiv:1802.09145 [nucl- ex]

    Acharya, S.,et al.: Transverse momentum spectra and nuclear modification fac- tors of charged particles in pp, p-Pb and Pb-Pb collisions at the LHC. JHEP11, 013 (2018) https://doi.org/10.1007/JHEP11(2018)013 arXiv:1802.09145 [nucl- ex]

    Pith/arXiv arXiv doi:10.1007/jhep11(2018)013 2018

  35. [36]

    Adam, J.,et al.: Pseudorapidity and transverse-momentum distributions of charged particles in proton–proton collisions at √s= 13 TeV. Phys. Lett. B753, 319–329 (2016) https://doi.org/10.1016/j.physletb.2015.12.030 arXiv:1509.08734 [nucl-ex] Appendix A Additional plots and tables Initialization is required to set up the calculations and the objects that th...

    Pith/arXiv arXiv doi:10.1016/j.physletb.2015.12.030 2016