---
id: 20260814-T0-05
title: "强化学习训练能耗实测：从单GPU到整机群的功率控制新方法"
title_en: "Paper: Reinforcement Learning Training Power Control Cuts AI Datacenter Energy from One GPU to Fleet"
url: https://ai.daily.yangsir.net/daily/20260814-T0-05
issue_date: 2026-08-14
publish_date: 2026-08-13T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2608.11226
---

# 强化学习训练能耗实测：从单GPU到整机群的功率控制新方法

一篇来自arXiv的新研究首次系统性地测量了强化学习后训练在GPU上的功耗行为，发现当前数据中心使用的静态功率上限和反应式节流机制效率低下。研究提出了一种基于强化学习的功率控制方法，在单个GPU和整个机群规模上都实现了实测能耗削减，为大规模LLM训练提供了可量化的节能方案。

## English Version

**Paper: Reinforcement Learning Training Power Control Cuts AI Datacenter Energy from One GPU to Fleet**

A new arXiv paper characterizes the previously unmeasured power behavior of RL post-training on GPUs, exposing weaknesses in workload-blind datacenter power management. The authors propose a reinforcement-learning-based power control method, showing measured energy reductions from single GPU to fleet level—a first for LLM training efficiency.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2608.11226)

**详情页**：https://ai.daily.yangsir.net/daily/20260814-T0-05

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*