ChatGPT 5 power consumption could be as much as eight times higher than GPT 4 — research institute estimates medium-sized GPT-5 response can consume up to 40 watt-hours of electricity
Technology
128
Beiträge
79
Kommentatoren
0
Aufrufe
-
Politicians are cheap
Yeah sorry forgot my /s there
-
The University of Rhode Island's AI lab estimates that GPT-5 averages just over 18 Wh per query, so putting all of ChatGPT's reported 2.5 billion requests a day through the model could see energy usage as high as 45 GWh.
A daily energy use of 45 GWh is enormous. A typical modern nuclear power plant produces between 1 and 1.6 GW of electricity per reactor per hour, so data centers running OpenAI's GPT-5 at 18 Wh per query could require the power equivalent of two to three nuclear power reactors, an amount that could be enough to power a small country.
that's a lot. remember to add "-noai" to your google searches.
-
Can you give some examples of those technologies? I'd be interested in how many weren't replaced with something more efficient or convenient.
There were certainly companies that survived, because yes, the idea of websites being interactive rather than informational was huge, but everyone jumped on that bandwagon to build useless shit.
As an example, this is today’s ProductHunt
And yesterday’s was AI, and the day before that it was AI, but most of them are demonstrating little value with high valuations.
LLMs will survive, likely improve into coordinator models that request data from SLMs and connect through MCP, but the investment bubble can’t sustain
-
that's a lot. remember to add "-noai" to your google searches.
I'm just going to ignore the AI recommendations, let them burn money.
-
It would only take one regulation to fix that:
Datacenters that use liquid cooling must use closed loop systems.
The reason they dont, and why they setup in the desert, is because water is incredibly cheap and energy to cool a closed loop system is expensive. So they use evaporative open loop systems.
That increases your energy use though, because evaporative cooling is very energy efficient.
-
I don't care how rough the estimate is, LLMs are using insane amounts of power, and the message I'm getting here is that the newest incarnation uses even more.
BTW a lot of it seems to be just inefficient coding as Deepseek has shown.
My guess would be that using a desktop computer to make the queries and read the results consumes more power than the LLM, at least in the case of quickly answering models.
The expensive part is training a model but usage is most likely not sold at a loss, so it can't use an unreasonable amount of energy.
Instead of this ridiculous energy argument, we should focus on the fact that AI (and other products that money is thrown at) aren't actually that useful but companies control the narrative. AI is particularly successful here with every CEO wanting in on it and people afraid it is so good it will end the world.
-
I'm just going to ignore the AI recommendations, let them burn money.
i don't judge you for that. honestly it matters fuck all at this point
-
Maybe you're mixing Wh with kWh. 40Wh is not that much, but it's still a lot for a single request.
Roughly the capacity of a laptop battery, a huge amount of energy per request.
-
That increases your energy use though, because evaporative cooling is very energy efficient.
We can make energy from renewable sources.
Fresh drinking water is finite, especially in the desert.
-
Tech hasn't improved that much in the last in the last decade. All that's happened is that more cores have been added. The single-thread speed of a CPU is stagnant.
My home PC consumes more power than my Pentium 3 consumed 25 years ago. All efficiency gains are lost to scaling for more processing power. All improvements in processing power are lost to shitty, bloated code.
We don't have the tech for AI. We're just scaling up to the electrical senand demand of a small country and pretending we have the tech for AI.
This is nonsense, an M1 runs many multiples faster and at much lower wattage.
-
The University of Rhode Island's AI lab estimates that GPT-5 averages just over 18 Wh per query, so putting all of ChatGPT's reported 2.5 billion requests a day through the model could see energy usage as high as 45 GWh.
A daily energy use of 45 GWh is enormous. A typical modern nuclear power plant produces between 1 and 1.6 GW of electricity per reactor per hour, so data centers running OpenAI's GPT-5 at 18 Wh per query could require the power equivalent of two to three nuclear power reactors, an amount that could be enough to power a small country.
Fucking Doc Brown could power a goddamn time machine with this many jiggawatts, fuck I hate being stuck in this timeline.
-
OpenAI are not profitable today, and don't estimate they'll be profitable until 2029, so it's almost guaranteed that they're selling their services at a loss. Of course, that's impossible to verify - since they're a private company, they don't have to release financial statements.
There's a difference between selling at a loss, and having a loss.
OpenAI let's people use models for free with very little limits other than reducing the model quality over time, and they have very generous limits before they limit you at that.
That all costs money and is a loss for them.
If they get someone who's willing to pay, and they charge $20/m and on average, they net $5 profit per customer, they aren't selling it at a loss, they just need more customers. It's possible that a paid customer uses it even more though and it actually does incur a loss per paid customer and they're doing that to try and gain users while they figure out how to lower their costs, but that seems less likely.
-
Coordinated SLM governors that can redirect queries to the appropriate SLM seems like a good solution.
That basically just sounds like Mixture of Experts
-
Well over the course of an hot or two, but it's correct that a dryer run even with heat pump is significantly more than 40wh
The original commenter said they mixed up wh and Kwh.
My entire house uses under 40Kwh a day, and it's winter where I am at the moment.
-
that's a lot. remember to add "-noai" to your google searches.
Or just use any other better search like Bing or duckduckgo. googol sucks and was never any good. Quit pushing ignorant garbage.
-
Roughly the capacity of a laptop battery, a huge amount of energy per request.
A very small laptop battery.
-
Or just use any other better search like Bing or duckduckgo. googol sucks and was never any good. Quit pushing ignorant garbage.
duckduckgo yes, but ... bing?
-
duckduckgo yes, but ... bing?
-
but they are merely doing it to defend AI.
No they're not, you can agree the research is garbage without defending AI. It literally assumes everything. GPT5 could be using eight times the power. It could be using half the power. It could be using a quadrillion times the power. Nobody knows, because they keep it secret.
Sorry, i meant the comments under the article on toms hardware. I see this on a every tech site: research says it’s impossible amount of power used, and everyone in the comments below will do half arsed calculations with other made up numbers, trying to disprove the researcher, and always only to defend and legitimise their use of ai.
FWIW I’m not saying that the research here is garbage; it’s a decent attempt estimating by at assuming what you don’t and cannot really know, as long as an OpenAI and their doesn’t publish the data. As far as I know only Mistral published anything at all. -
The University of Rhode Island's AI lab estimates that GPT-5 averages just over 18 Wh per query, so putting all of ChatGPT's reported 2.5 billion requests a day through the model could see energy usage as high as 45 GWh.
A daily energy use of 45 GWh is enormous. A typical modern nuclear power plant produces between 1 and 1.6 GW of electricity per reactor per hour, so data centers running OpenAI's GPT-5 at 18 Wh per query could require the power equivalent of two to three nuclear power reactors, an amount that could be enough to power a small country.
The team measured GPT-5’s power consumption by combining two key factors: how long the model took to respond to a given request, and the estimated average power draw of the hardware [they believe is] running it.
