Become a fan of Slashdot on Facebook

 


Forgot your password?
Close
typodupeerror
×
AI

AI Computing Is on Pace To Consume More Energy Than India, Arm Says (yahoo.com) 50

Posted by msmash from the closer-look dept.
AI's voracious need for computing power is threatening to overwhelm energy sources, requiring the industry to change its approach to the technology, according to Arm Chief Executive Officer Rene Haas. From a report: By 2030, the world's data centers are on course to use more electricity than India, the world's most populous country, Haas said. Finding ways to head off that projected tripling of energy use is paramount if artificial intelligence is going to achieve its promise, he said.

"We are still incredibly in the early days in terms of the capabilities," Haas said in an interview. For AI systems to get better, they will need more training -- a stage that involves bombarding the software with data -- and that's going to run up against the limits of energy capacity, he said.
This discussion has been archived. No new comments can be posted.

AI Computing Is on Pace To Consume More Energy Than India, Arm Says More

AI Computing Is on Pace To Consume More Energy Than India, Arm Says

Comments Filter:

  • Sigh... (Score:3, Funny)

    by tonytins ( 10331799 ) on Wednesday April 17, 2024 @02:54PM (#64402058)
    Here we go again.

    • Re: (Score:2)

      by Rei ( 128717 )

      Here we go again with this [xkcd.com].

      NVidia shipped 100k AI GPUs last year, which - if run nonstop - would consume 7,4 TWh. Crypto consumes over 100 TWh per year, and the world as a whole consumes just under 25000 TWh per year.

      AI consumption of power is a pittiance. To get these huge numbers, they have to assume long-term extreme exponential scaling. But you can make anything give insane numbers with an assumption like that.

      I simply don't buy the assumption. Not even assuming an AI bust - even assuming that AI ke

      • Re: (Score:3)

        by ls671 ( 1122017 )

        Anyway, in The Matrix, everything was done for much cheaper with human brains, maybe we should take note of that!

      • Modern crypto mining uses hundreds, if not thousands, of small dedicated machines with an application-specific integrated circuit (ASIC). But before crypto reached its peak, GPUs were also thought to be the most efficient method. Obviously, they were wrong. However, even with ASICs, crypto still consumes tones of power. AI is still in its infancy and right now, it is consuming a lot of power. Just because it isn't to crypto's scale right now, doesn't mean we shouldn't address the issue while we still have t
  • Get to work on that Dyson Sphere. That should be enough.

    • That's not how exponential growth works.

      The only thing that can grow without bound is space itself. To everything else, there's limits. Even to the sun.

      OK, I guess we'll run out of materials by then - even to construct a Dyson sphere, there isn't nearly enough stuff even if one collected every rock floating around in the solar system.

    • Power satellites dedicated to AI are a thought. Heat sinks are a problem, but there is a 1.5 GW heat sink about 5 minutes into this video.https://www.youtube.com/watch?v=VEkZkINrJaA

  • Obviously the companies will go wherever profits indicate, but memristor tech is theoretically a massive leap forward, allowing low-power transistor equivalents, using less volume, maintaining persistent states without power, and providing more than just binary states per memristor. And it's been in labs for at least a decade.

    Basically, if you're trying to make a neural net in hardware, memristors appear to be what you want.

  • Comment removed based on user account deletion

    • Renewable energy is not a free lunch. The sun doesnt shine at night, the wind doesnt always blow and hydro can run out of water in a drought and there isnt enough of it for current needs never mind projected ones including the massive rise from EV charging.

      • That's why you combine generation and storage (and note I didn't say "batteries"). That storage can include turning atmospheric co2 into fuel.

        But underlying your point is that we simply have too many people. Generating baseline power for the current population is rendering the planet uninhabitable.

      • ROFL.. you worried about EV charging? I'd be more worried about the energy use referred to in the article. At least when V2G is available then EVs can help the grid at times of stress.
    • Nope most of what you think is renewable energy is all created by fossil fuels and the maintenance and parts of renewable sources need fossil fuels
      • ah.. another one who doesn't understand how a power transition works - hint: you have to use the old tech to create the new unless you are a wizard and can magic things into existence.

    • Re: (Score:1)

      by Tablizer ( 95088 )

      Ironically AI may help us finally make fusion energy feasible. I hope...

  • Let's generate still more greenhouse gases, while simultaneously dumping more heat into our atmosphere, in the cause of corporations firing more people and wealth being concentrated further still. Then we can move faster toward the dystopia outlined in the film Elysium. Historians may look back on all of this and call it The Grand Cost Externalization - assuming that civilization survives long enough for said historians to be born.

    Is it time yet to break out the torches and pitchforks?

  • This is kind of interesting, but I'm also thinking WTF?

    What kind of terrible code do you have that ingests petabytes of data including billions of pages of literature and can't come up with something that can beat a young adult human who's read tiny fractions of that?

    • Yeah if that's how much energy it takes to brute force the output of an administrative assistant or technical writer than surely the latter are cheaper

    • Massively parallel vector multiplication. Currently big companies are eating the energy bills because they're stuck in the upward portion of a hype cycle, but soon they'll have to start charging people to talk to their chat bots. They don't get energy for free either.

  • Seems to me that focusing on increased efficiency would have huge payoffs. Brains do this with a very tiny amount of electricity so the potential power efficiencies are vast.

  • how long till we are combined with a certain type of fusion so the AI will have all the energy it needs??

  • A long time ago, people realized progress depended on energy. If you have enough energy, you can do almost anything.

    Best to get to it. If it needs to be "clean", so be it, but conservation AKA Spartan lives, is of very limited conceptual value.

    Get to it making more.

  • is humans. The average human body generates more bio-electricity than a 120-volt battery and over 25,000 BTUs of body heat.

  • At least we know that the LLM will not hallucinate the em-phaasis on the wrong sylll-ahble. They're at least good at that part.

    But if we're trying to put India out of business and replace them with AI, this doesn't seem good: India will consume less energy doing the same thing as the AI!

  • By 2030, the world’s data centers are on course to use more electricity than India, the world’s most populous country, Haas said.

    "The world's data centers" encompass a whole lot more than just *AI Computing*.

  • Currently data centers consume about 1% of worldwide electricity, and India consumes about 6%. However, we're talking about 6-year projections. How much will India use in 6 years, and how much will data centers use? Increasing data center electricity usage by 6x sounds like a lot, but increasing to 6% of total usage sounds less ominous.

    Comparing to India is an attempt to place the issue in a context that is understandable for everyone. If so, that was a fail, since almost no one has a good feel for how much

  • They could always build nuclear power plants and have more electricity than we could ever productively use... but nah. Let's burn some more coal.

Slashdot Top Deals

16.5 feet in the Twilight Zone = 1 Rod Serling

Close