“Your comment is addressed in the introduction section of the paper being discussed. Do give it a read.”
Making AI Less "Thirsty": Uncovering and Addressing the Secret Water Footprint of AI Models
提出估算 AI 水足迹的方法,量化训练与推理耗水
论文指出,AI 的碳足迹已受到关注,但同样重要的水足迹(取水与耗水)一直被忽视,例如在微软美国数据中心训练 GPT-3 就可直接蒸发 70 万升淡水,并预计 2027 年全球 AI 需求的取水量达 42–66 亿立方米。作者给出估算 AI 水足迹的原则性方法,讨论运行时水效率在时间和空间上的差异,并呼吁将水足迹与碳足迹一并考虑。
每当有文章讨论 AI 用水量,评论者就拿它当数据来源,例如大部分用水来自发电而非冷却、单次推理约耗 10–50 毫升;也有人指出其估算基于 GPT-3 时代的旧数字。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及“comes from an April 2023 paper (Li et al, arXiv:2304.03271) that estimates water/power usage based off of GPT-3 (175B dense model) numbers published from OpenAI's original GPT-3 2021 paper.”
“The paper[1] mentions the statistics used to calculate these numbers.”
“Link to the paper in question: Making AI Less "Thirsty"”
“According to the paper, the vast majority of water usage is not from cooling but indirectly from energy usage.”
“[1]: Using same electricity water intensity as used in Making AI Less "Thirsty"”
看全部 9 条 →
“If you look at the original paper they are quite upfront with the difficulty of estimating water use.”
“which is an order of magnitude larger than any figure in that paper”
“According to this study[1]”
- Uncovering and Addressing the Secret Water Footprint of AI Models ↗1vuio0pswjnm7 · 2024-03-11 · 1 分
- Secret Water Footprint of AI Models ↗raptorraver · 2024-08-11 · 1 分
- "GPT training in U.S. data centers can evaporate 700k L of freshwater" ↗TaurenHunter · 2025-01-15 · 3 分
- Making AI Less Thirsty: Uncovering and Addressing the Secret AI Water Footprint ↗eicnix · 2025-03-04 · 2 分
- Uncovering and addressing the secret water footprint of AI models ↗1vuio0pswjnm7 · 2025-06-29 · 1 分