大模型
arxiv.org ·
面向功率灵活型 AI 训练的作业功率弹性表征
该论文对 LLM 训练的“作业功率弹性”（吞吐量对功率削减的敏感度）进行首次系统性表征，提出归一化指标 Power Flexibility Index（PFI）作为 SLA 感知的功率灵活性控制原语。作者采集 131 次 H200 训练运行数据（另含 24 次 H200 验证与 34 次 H100 对照），覆盖 dense 与 MoE 模型、预训练与微调任务、最多 32 块 GPU。结果显示训练作业功率弹性显著但差异较大，并识别出可在运行时预测 PFI 的遥测信号；PFI 感知的功率分配可在功率约束下最大化总 tokens/s 吞吐，在 30% 功率削减下每作业恢复约 1.5k tokens/s，相当于等权分配与完美信息 oracle 性能差距的 63%。
AI推荐理由 arXiv 新论文，提出首个 LLM 训练作业功率弹性（job power elasticity）的系统性表征，并给出可量化的 Power Flexibility Index（PFI）。基于 131 次 H200 训练运行（另含 24 次 H200 验证与 34 次 H100 对照），覆盖 dense 与 MoE、预训练与微调、最多 32 GPU，数据规模与实验设计具体可查。核心结论是训练作业功率弹性显著但差异大，且可用运行时遥测信号预测 PFI；在 30% 功率削减下，PFI 感知分配每作业可恢复约 1.5k tokens/s，达到等权分配与完美信息 oracle 之间性能差距的 63%。对 AI 基础设施、数据中心电力调度与算力落地有直接参考价值。局限在于结论基于特定硬件与最多 32 GPU 规模，向更大集群外推需谨慎；摘要未给出完整方法细节与统计显著性。
82
正文
Computer Science > Artificial Intelligence
arXiv:2609.11542 (cs)
[Submitted on 10 Sep 2026]
Title:Characterizing Job Power Elasticity for Power-Flexible AI Training
Authors:Philip Colangelo, Charles Dawson, Shayan Sengupta, Ayse Coskun, Varun Sivaram
View a PDF of the paper titled Characterizing Job Power Elasticity for Power-Flexible AI Training, by Philip Colangelo and 4 other authors
View PDF
HTML (experimental)
Abstract:Large language model (LLM) training is among the fastest-growing sources of electricity demand in modern data centers, and power availability is a primary bottleneck to continued AI infrastructure growth. Making the power consumption of these workloads flexible could unlock additional power for AI growth, limit increases in electricity prices, and improve the utilization of existing grid infrastructure. However, to realize this flexibility, we must first understand how the performance of training workloads changes when GPU power is reduced.
This paper presents the first systematic characterization of \emph{job power elasticity} (the sensitivity of throughput to power reductions) in LLM training. To quantify elasticity, we introduce the \emph{Power Flexibility Index (PFI)}, a normalized metric that quantifies the performance cost of power reductions and provides a control primitive for SLA-aware power flexibility.
We collect data from 131 LLM training runs on H200 (plus 24 H200 validation runs and 34 matched H100 runs), including both dense and mixture-of-experts models, pretraining and fine-tuning tasks, and up to 32 GPUs. We find that LLM training jobs exhibit substantial but variable power elasticity, and we identify telemetry signals that predict PFI at runtime. Finally, we demonstrate that PFI-aware power allocation maximizes total tokens/second throughput under power constraints. Under a 30\% power reduction, PFI-aware power allocation recovers ~1.5k tokens/s per job, 63\% of the performance gap between an equal-weight allocation and an oracle with perfect information. Our results establish power elasticity as a measurable property of training jobs and provide a foundation for power-aware, grid-responsive AI infrastructure.
Subjects:
Artificial Intelligence (cs.AI)
Cite as:
arXiv:2609.11542 [cs.AI]
(or
arXiv:2609.11542v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.11542
Focus to learn more
arXiv-issued DOI via DataCite (pending registration)
Submission history
From: Charles Dawson [view email]
[v1]
Thu, 10 Sep 2026 13:40:13 UTC (2,136 KB)
Full-text links:
Access Paper:
View a PDF of the paper titled Characterizing Job Power Elasticity for Power-Flexible AI Training, by Philip Colangelo and 4 other authors
View PDF
HTML (experimental)
TeX Source
view license
Current browse context:
cs.AI
new
|
recent
| 2026-09
Change to browse by:
cs
References & Citations
NASA ADS
Google Scholar
Semantic Scholar
export BibTeX citation
Loading...
BibTeX formatted citation
×
loading...
Data provided by:
Bookmark
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media
Code, Data and Media Associated with this Article
alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos
Demos
Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers
Recommenders and Search Tools
Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
Author
Venue
Institution
Topic
About arXivLabs
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Which authors of this paper are endorsers? |
Disable MathJax (What is MathJax?)