One million Blackwell GPUs would suck down an astonishing 1.875 gigawatts of power. For context, a typical nuclear power plant only produces 1 gigawatt of power.
Fossil fuel-burning plants, whether that’s natural gas, coal, or oil, produce even less. There’s no way to ramp up nuclear capacity in the time it will take to supply these millions of chips, so much, if not all, of that extra power demand is going to come from carbon-emitting sources.
NVidia designing, building and selling these sorts of cards with astronomical power usage? I get it. They want to stay at the top.
But those buying these cards at least need to be taxed, charged, regulated, whatever to make sure the huge additional power they require is funded by said company, and should only be green/renewable energy sources. And not using clean drinking water communities need for cooling.
If companies want to run massive amounts of hardware like this, it should be prohibitively expensive unless they build their own GREEN power stations, and find ways to cool without using drinking water from any community.
At the moment, taxes and government money goes into power stations which these DCs then use. All the cost is pushed right down onto the every day tax payer and consumer. But all the profit is flowing upwards.
Make them pay for what they use. Make them pay to make these cards efficient, clean, and safe for our environment. Its not like these trillion dollar companies couldn’t pay for it all and make the world a better place.
Using tons of energy isn’t a problem, as long as it’s carbon neutral (or negative). The problem is that we are simply not there yet. Taxing carbon is a great solution and would nearly immediately fix the problem (on the scale of years, not decades).
Using tons of energy isn’t a problem, as long as it’s carbon neutral (or negative).
That energy should go to more useful-to-society purposes, first. If all the “AI” datacenters are running on green power and the rest of us are still burning coal, then that’s green power that’s still wasted. It’s even more of a slap in the face if taxpayer funds go toward the costs of building any of those single-purpose green energy projects.
Bingo, same argument I’ve had to bring up with crypto bros, if it’s using green energy that could be used to power essential things that are powered by fossil fuel then it’s green energy that we’re wasting. All these projects should be put on hold until we’re running on 100% green energy and we produce enough surplus that we can afford to use it for non essentials.
You only need 1.21 gigawatts to go back to the future in a Delorean.
I think you mean 1 point 21 jigowatts
“I always figured that the word “Jigawatt”was made up just for the movie and meant to sound like a really large amount. It wasn’t unit I started researching this flux capacitor replica project that I stumbled across a few references to the actual term. It turns out that the original pronunciation of “Giga” was with the “j” sound (really a soft “g”).” Here Check out Merriam Webster’s pronunciation
I’ve had multiple people on Lemmy tell me that the amount of energy LLMs use will be trivial. They always base it on the amount of energy used to train the LLMs, not the millions (billions? trillions?) of calculations those LLMs have to do every second they’re used by who knows how many people 24 hours a day.
Then you bring up the water wasting and the best they can do is say something like, “okay, that’s a problem… but only in some places!”
(Some places including much of the United States. Guess where lots of the data centers are?)
I don’t disagree, but it is useful to point out there are two truths in what you wrote.
The energy use of one person running an already trained model on their own hardware is trivial.
Even the energy use of many many people using already trained models (ChatGPT, etc) is still not the problem at hand (probably on the order of the energy usage from a typical search engine).
The energy use in training these models (the appendage measuring contest between tech giants pretending they’re on the cusp of AGI) is where the cost really ramps up.
(probably on the order of the energy usage from a typical search engine).
I find that hard to believe. Search engines just regurgitate what is in a database. LLMs have to do calculations to create the sentences they produce. That takes more energy.
Believe what you will. I’m not an authority on the topic, but as a researcher in an adjacent field I have a pretty good idea. I also self host Ollama and SearXNG (a metasearch engine, to be clear, not a first party search engine) so I have some anecdotal inclinations.
Training even a teeny tiny LLM or ML model can run a typical gaming desktop at 100% for days. Sending a query to a pretrained model hardly even shows up on HTop unless it’s gigantic. Even the gigantic models only spike the CPU for a few seconds (until the query is complete). SearXNG, again anecdotally, spikes my PC about the same as Mistral in Ollama.
I would encourage you to look at more explanations like the one below. I’m not just blowing smoke, and I’m not dismissing the very real problem of massive training costs (in money, energy, and water) that you’re pointing out.
https://www.baeldung.com/cs/chatgpt-large-language-models-power-consumption
