← 論文一覧に戻る

高性能計算のカーボンフットプリント測定の再現パッケージ：炭素会計の選択が答えをどう変えるか

Reproducibility Package for Measuring the Carbon Footprint of High-Performance Computing: How Carbon Accounting Choices Change the Answer (原題)

Anonymous, Anonymous

Zenodo (CERN European Organization for Nuclear Research)ジャーナル2026-11-01#炭素会計Origin: US対象セクター: cross_sector
DOI: 10.5281/zenodo.23851464
原典: https://doi.org/10.5281/zenodo.23851464

🤖 gxceed AI 要約

日本語

HPCの運用炭素会計における手法選択（電力グリッド炭素強度の基準、時間解像度、PUE処理）が報告フットプリントとワークロード帰属に与える影響を、Kestrelの370万ジョブ・7.437GWhのデータで検証。再現パッケージとしてプロトコル、ノートブック、派生表、監査マニフェストを提供。会計選択の感度と転用可能性を評価する。

English

Examines how methodological choices in operational carbon accounting (grid carbon-intensity basis, temporal resolution, PUE treatment) affect reported carbon footprint and workload attribution for HPC, using 3.7M Kestrel jobs (7.437 GWh). Provides a reproducibility package with protocol, notebooks, derived tables, and audit manifests. Includes a cross-balancing-authority signal-substitution stress test of accounting transferability.

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

📝 gxceed 編集解説 — Why this matters

日本のGX文脈において

SSBJや有報でのScope 2・3算定において、電力炭素強度の選択や時間解像度が結果を大きく変えうることを示す。データセンターや研究機関の排出量開示の精度向上に寄与し、日本企業のGHG算定ガイドライン策定にも示唆を与える。

In the global GX context

Highlights the sensitivity of Scope 2/3 accounting to methodological choices (grid intensity, temporal resolution, PUE), relevant to ISSB/CSRD disclosure quality and assurance. Provides a reproducible framework for testing accounting transferability across balancing authorities, supporting more robust corporate GHG inventories.

👥 読者別の含意

🔬研究者:炭素会計の方法論的選択が結果に与える影響を定量化する再現可能な枠組みを提供。

🏢実務担当者:データセンターやHPC利用企業がScope 2/3算定の感度を理解し、開示の頑健性を高めるのに役立つ。

🏛政策担当者:GHG算定ガイドラインにおける電力炭素強度や時間解像度の規定の重要性を示唆。

📄 Abstract（原文）

This record provides the reproducibility package for the study “Measuring the Carbon Footprint of High-Performance Computing: How Carbon Accounting Choices Change the Answer.” The study examines how methodological choices in operational carbon accounting affect the reported carbon footprint and workload-level attribution of high-performance computing (HPC). The primary analysis uses 3,707,342 completed, non-GPU, full-node Kestrel jobs representing 7.437 GWh of Slurm-reported IT energy. It compares alternative electricity-grid carbon-intensity bases, temporal resolutions, and facility PUE treatments while holding the observed workloads and their historical execution times fixed. The archive contains: the frozen study protocol, decision log, and integrity/audit manifests; Jupyter/Google Colab notebooks implementing the data preparation, carbon-accounting, bootstrap, robustness, and decision-stability analyses; compact derived tables and statistical outputs supporting the manuscript; post-freeze exploratory analyses of EPA eGRID reporting factors, account-level aggregation, and cross-balancing-authority transferability; manuscript table and figure source data; and SHA-256 file-integrity information and citation metadata. The cross-balancing-authority analysis holds the Kestrel workload, execution times, energy allocation, and facility treatment fixed while substituting hourly EIA-930 signals from PSCO, CISO, NYIS, ERCO, and PJM. It is a signal-substitution stress test of accounting transferability and is not a geographic relocation or scheduling counterfactual. Large public source datasets and large regenerable intermediate files are not redistributed in this archive. The included README identifies the source datasets and describes the workflow required to regenerate the analysis. This study concerns historical attribution of operational electricity-related CO2. It does not estimate emissions savings from workload rescheduling, deferral, geographic relocation, packing, or other counterfactual interventions.

🔗 Provenance — このレコードを発見したソース

🔔 こうした論文の新着を逃したくない方は キーワードアラート に登録（無料・3キーワードまで）。

gxceed は公開メタデータに基づく研究支援データセットです。要約・翻訳・解説は AI 支援で生成されています。 最終的な解釈・検証は利用者が原典資料に基づいて行うことを前提とします。