#EnergyConsumption

PUPUWEB Blogpupuweb
2025-06-10

💡 Sam Altman reveals ChatGPT energy use: 0.34 watt-hours per query (1 second of oven use) + 0.000085 gallons water (1/15th teaspoon). The true cost of AI conversations 📊

N-gated Hacker Newsngate
2025-06-05

🔥 Oh joy, another “revolution” in AI where we pretend our computers are brains, except they're not, and they never will be. 🔋💸 Let's burn through enough electricity to power a small galaxy and then claim we're saving energy. 🌌💡 Neuromorphic computing: when marketing meets science fiction. 🤖🙄
lanl.gov/media/publications/16

2025-06-04

MIT Tech Review's AI Energy Package uncovers the immense energy and water consumption involved in training and operating AI models. This crucial read sheds light on the unseen environmental impact of AI and its implications for a sustainable technological future.

Explore more: technologyreview.com/supertopi

#AI #Sustainability #EnergyConsumption

Eugenus Optimus 🇺🇦ujeenator@ujeenator.net
2025-05-21

https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/

🔸 MIT Technology Review provided a comprehensive per-query energy analysis for AI text, image, and video models, accounting for hidden infrastructure and cooling costs.
🔸 AI’s energy demand is rapidly escalating – Meta and Microsoft are eyeing new nuclear plants; OpenAI’s Stargate initiative proposes $500 billion for up to 10 data centers; Apple plans $500 billion in US manufacturing and data centers; Google will spend $75 billion on AI infrastructure in 2025.
🔸 From 2017 to 2023, data-center electricity consumption doubled, now consuming 4.4 % of US power, with a carbon intensity 48 % higher than the US average.
🔸 Inference (serving user queries) consumes 80–90 % of AI’s computing power, surpassing training workloads.
🔸 Text-model inference energy – Meta Llama 3.1 8B uses ~57 J GPU (114 J total) per response; Llama 3.1 405B uses ~3,353 J GPU (6,706 J total).
🔸 Image generation – Stable Diffusion 3 Medium (1024×1024) uses ~1,141 J GPU (2,282 J total); doubling diffusion steps roughly doubles energy.
🔸 Video generation – CogVideoX’s older model used ~109 kJ per 5 s clip; its newer higher-quality version uses ~3.4 MJ per 5 s.
🔸 A hypothetical day of 15 text queries, 10 image attempts, and 3 five-second videos consumes ~2.9 kWh (≈10 minutes of microwave or 100+ miles on an e-bike).
🔸 Data centers rely on ~60 % fossil fuels, 20 % nuclear, 20 % renewables – often tapping dirtier grids that run 24/7.
🔸 Carbon intensity varies by region and time (e.g., California 70 gCO₂/kWh by day vs 300 gCO₂/kWh at night; West Virginia much higher).
🔸 In 2024 US data centers used ≈200 TWh; AI-specific servers ≈53–76 TWh; by 2028 AI could demand 165–326 TWh/year, enough for 22 % of US households.
🔸 Closed-source model providers share little data, creating “black boxes” that impede accurate energy and emissions planning.

#ai #machinelearning #datacenters #climatechange #energyconsumption #carbonemissions #sustainability #renewableenergy #nuclearenergy #technology #bigtech #openai #google #microsoft #meta

2025-05-20

#AI's energy footprint is growing fast; each #ChatGPT query may use 0.3Wh, adding up to over 109GWh annually for a billion daily queries! Adding images boosts that even more. How will AI scale #sustainably?
#AI #EnergyConsumption #Sustainability

Kathy BrysonKathyBryson
2025-04-29

An important question - do you need to be polite to your

Sam Altman Admits Saying "Please" and "Thank You" to ChatGPT Wastes Millions in Computing Power buff.ly/KBKCSUm

JustAGuyJustAGuy2010
2025-04-23

🔥⚡ Elon Musk’s AI Ambitions Spark Environmental Concerns

Elon Musk’s AI company, xAI, is under scrutiny for operating 35 unpermitted methane gas-burning generators at its Memphis facility, generating approximately 420MW of electricity—enough to power an entire city. Community advocates and environmental groups accuse xAI of violating the Clean Air Act.

👉 theguardian.com/us-news/2025/a

Thomas Fricke (he/him)thomasfricke@23.social
2025-04-13

Dirtiest US coal-fired power plant applies for EPA exemption

Colstrip power plant, Montana. Image source: The Center for Land Use Interpretation

The dirtiest coal-fired power plant in the US is asking Donald Trump for a waiver from pollution mandates, taking up the administration on an offer to email for a chance to get a presidential exemption..

mining.com/web/dirtiest-us-coa

#AI #datacenter #energyconsumption

Technische Universität Münchentu_muenchen@wisskomm.social
2025-04-10

At our new energy laboratory, technology and #sustainable #energysystems become tangible. Schoolchildren can observe the #energyconsumption of household appliances in real time: go.tum.de/483742 🔌

#energytransition

📷J.Forstner

Thomas Fricke (he/him)thomasfricke@23.social
2025-04-06

Gutes Interview von @tazgetroete mit @Rainer_Rehak von @bitsundbaeume_berlin und @Weizenbaum_Institut

Künstliche Intelligenz: Was wir gegen den massiven Stromverbrauch tun können | taz.de
taz.de/Stromfresser-Kuenstlich

#ai #datacenter #energyconsumption

2025-03-25

🤖 ⚡ It takes ten times more electricity for ChatGPT to respond to a prompt than for Google to carry out a standard search. Still, researchers are struggling to get a grasp on the energy implications of generative artificial intelligence both now and going forward.

#ChatGPT #GenerativeAI #EnergyConsumption

Read more: go.epfl.ch/bgA-en

Kevin Karhan :verified:kkarhan@infosec.space
2025-03-14

@fuchsiii @grumpygamer oftentimes said cores ain't even "dedicaded" but shared...

  • In the case of #CyberBunker they literally sold #Nehalem(-EX) hardware to customers (cross referenced by datasheets/specsheets and the fact they never updated the listings in 15+ years!)

"shared core VPS" are a very big mistake, espechally on Hosters like #Hetzner...

  • Ask me how I know!

So OFC said systems are overworked, espechally the convenience-based options where one just shoves on files and tells it to "bake it"...

  • OFC it shouldn't be the responsibility of a #GameDev to #DIY one's own #CI / #CD infrastructure, and I think given the size and complexity of said games it would be overkill to do so.

If feasible, just grab' yourself a decently performing PC/Server/Laptop you can chug in a closet, SSH/RDP onto and use it as "build servant" ("Rechenknecht") to run your builds and then sent a notification once done.

Thomas Fricke (he/him)thomasfricke@23.social
2025-03-14

In #Brandenburg bei #Nauen wird ein 200 MW #datacenter gebaut. Die Integration klingt erstmal gut. Erneuerbare, Energiespeicher Anschluss an Fernwärme.

Ich hoffe, sie haben auch den Wasserverbrauch im Griff. Eigentlich gibt es da genug.

datacenter-insider.de/maincube

#AI #energyconsumption #renewables

Technische Universität Münchentu_muenchen@wisskomm.social
2025-03-06

To train #neuralnetworks more efficiently and reduce #energyconsumption, researchers developed a method that directly calculates parameters based on probabilities, rather than using an iterative approach: go.tum.de/972782

#AI #LLMs #MachineLearning

📷V.Hohenegger / LRZ

@c_9 this is a known ecological issue but no one wants to respond to comments or questions about the extraordinary amount of energy AI consumes to produce what it produces.

#AI #energyconsumption

Client Info

Server: https://mastodon.social
Version: 2025.04
Repository: https://github.com/cyevgeniy/lmst