frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

Open in hackernews

Ask HN: How expensive are LLMs to query, really?

5•teach•1d ago
I'm starting to see things pop-up from well-meaning people worried about the environmental cost of large language models. Just yesterday I saw a meme on social media that suggested that "ChatGPT uses 1-3 bottles of water for cooling for every query you put into it."

This seems unlikely to me, but what is the truth?

I understand that _training_ an LLM is very very expensive. (Although so is spinning up a fab for a new CPU.) But it seems to me the incremental costs to query a model should be relatively low.

I'd love to see your back-of-the-envelope calculations for how much water and especially how much electricity it takes to "answer a single query" from, say, ChatGPT, Claude-3.7-Sonnet or Gemini Flash. Bonus points if you compare it to watching five minutes of a YouTube video or doing a Google search.

Links to sources would also be appreciated.

Comments

serendipty01•1d ago
Some links:

https://www.sustainabilitybynumbers.com/p/carbon-footprint-c...

https://andymasley.substack.com/p/a-cheat-sheet-for-conversa...

(discussion on lobste.rs - https://lobste.rs/s/bxixuu/cheat_sheet_for_why_using_chatgpt...)

(discussion on HN, 320 comments: https://news.ycombinator.com/item?id=42745847)

teach•1d ago
These are excellent, thank you!
a_conservative•1d ago
my m4max macbook can run local inference on a medium-ish gemini model (32b IIRC). The power consumption spikes by about 120 watts over idle (with multiple electron apps, docker, etc). It runs about 70 tokens/sec and usually responds within 10 to 20 seconds.

So.. picking some numbers for calculation. 4 answers per minute @ 120 watts is about .5 watt-hours per answer. ~200 responses would be enough to drain the (normally quite long lasting battery).

How does that compare to the more common nvidia GPUs? I don't know.

The Universe of Discourse: A puzzle about balancing test tubes in a centrifuge

https://blog.plover.com/math/centrifuge.html
1•pavel_lishin•35s ago•0 comments

High-School Shop Students Attract Skilled-Trades Job Offers

https://www.wsj.com/lifestyle/careers/skilled-trades-high-school-recruitment-fd9f8257
2•lxm•5m ago•0 comments

FUTOcore: A New Software Store

https://www.youtube.com/watch?v=JMSGOVWKn80
1•pjmlp•6m ago•0 comments

Everel single-blade propeller [1938, 2018]

https://www.aopa.org/news-and-media/all-news/2018/november/pilot/singular-sensation
1•lisper•7m ago•0 comments

Tell HN: I use AI to help me code, but I don't want to be called a "vibe coder"

2•owendarko•7m ago•0 comments

Immunogenicity and Safety of Influenza and Covid-19 Multicomponent Vaccine

https://jamanetwork.com/journals/jama/article-abstract/2833668
1•bikenaga•8m ago•0 comments

DNS Piracy Blocking Orders: Google, Cloudflare, and OpenDNS Respond Differently

https://torrentfreak.com/dns-piracy-blocking-orders-google-cloudflare-and-opendns-respond-differently-250511/
2•DanAtC•10m ago•0 comments

HunyuanVideo-I2V: 14B model turns an image into 720p video on 8GB GPU

https://wavespeed.ai/models/wavespeed-ai/hunyuan-video/i2v
2•sylm•13m ago•1 comments

Authorization Code Flow for Server-Side Apps

https://developer.yahoo.com/oauth2/guide/flows_authcode/
2•mooreds•15m ago•0 comments

Tech that defined the modern internet is changing, SV is finally admitting it

https://www.cnn.com/2025/05/11/tech/google-facebook-silicon-valley-changes
1•lwo32k•16m ago•0 comments

I'm becoming increasingly worried about AI (2017)

https://www.econlib.org/archives/2017/03/im_becoming_inc.html
1•mooreds•16m ago•0 comments

Trends in Educational Attainment in the U.S. Labor Force

https://www.calculatedriskblog.com/2025/05/trends-in-educational-attainment-in-us.html
1•mooreds•17m ago•0 comments

First time founders are obsessed with product. 2nd time worry about distribution

https://twitter.com/justinkan/status/1418003365695418373
1•sanj•18m ago•0 comments

Global emergence of unprecedented lifetime exposure to climate extremes

https://www.nature.com/articles/s41586-025-08907-1
1•rntn•22m ago•0 comments

What Is Programming?

https://cacm.acm.org/opinion/what-is-programming/
1•joaogui1•22m ago•0 comments

Plotting Truth vs. Predicted Value

https://statmodeling.stat.columbia.edu/2025/05/11/plotting-truth-vs-predicted-value/
1•Tomte•23m ago•0 comments

Show HN: One-liner CLI for batched PDF-to-Markdown at $1 per ~6k pages

https://github.com/altaidevorg/llm-food
3•monatis•25m ago•0 comments

Feelings, Facts, and Our Crisis of Truth

https://thedispatch.com/article/feelings-facts-and-our-crisis-of-truth/
1•furrowedbrow•25m ago•0 comments

A close reading of the AI fake cases judgement

https://davidallengreen.com/2025/05/a-close-reading-of-the-ai-fake-cases-judgment/
2•mike-the-mikado•27m ago•1 comments

Show HN: LLM Agents Play Among Us-Like Game

https://github.com/The-Pocket/PocketFlow-Tutorial-Danganronpa-Simulator
1•zh2408•31m ago•0 comments

Family creates AI video to depict Arizona man addressing his killer in court

https://www.reuters.com/business/media-telecom/family-creates-ai-video-depict-arizona-man-addressing-his-killer-court-2025-05-09/
1•Anon84•34m ago•0 comments

Antarctica's Astonishing Rebound: Ice Sheet Grows in Decades

https://scitechdaily.com/antarcticas-astonishing-rebound-ice-sheet-grows-for-the-first-time-in-decades/
1•ksec•37m ago•0 comments

MSG Is (Once Again) Back on the Table

https://www.wired.com/story/msg-is-back/
1•ecliptik•39m ago•1 comments

How do builders solve the distribution problem in a world of daily launches?

https://old.reddit.com/r/ycombinator/comments/1kk0qp7/how_do_earlystage_builders_solve_the_distribution/
1•wslh•41m ago•0 comments

Gonzalo Guerrero

https://en.wikipedia.org/wiki/Gonzalo_Guerrero
4•akkartik•42m ago•0 comments

Manage the most important part of resource to save time and improve productivity

https://metaesn.com/
1•lilerjee•43m ago•1 comments

Would a plug-and-play abuse protection toolkit be useful beyond Stripe Radar?

1•MrDotNobi•44m ago•0 comments

2025 will likely be another brutal year of failed startups, data suggests

https://techcrunch.com/2025/01/26/2025-will-likely-be-another-brutal-year-of-failed-startups-data-suggests/
4•Bluestein•46m ago•0 comments

Wearable continuous diffusion-based skin gas analysis

https://www.nature.com/articles/s41467-025-59629-x
1•bookofjoe•47m ago•0 comments

Incus: System container and virtual machine manager

https://github.com/lxc/incus
1•tanelpoder•49m ago•0 comments