Comments

• Perspectivist@feddit.uk ⁨2⁩ ⁨days⁩ ago

  It’s a false assumption that AI bubble popping means we suddenly stop using AI and all these data centres will be running idle.

  We didn’t stop using the internet when the dotcom bubble burst either.

  AI bubble is about companies being overvalued. Not about the underlaying technology being worthless. The bubble popping means these values get adjustest to more reasonable levels. Some companies will go under but not all of them and almost definitely not the big ones.

  source

  • iloveDigit@piefed.social ⁨2⁩ ⁨days⁩ ago

    A small one would still be plenty of GPU

    And the big ones could still need to liquidate old stuff when it starts using too much power compared to newer stuff, even if it still works (though someone in this thread claimed data centers burn through GPUs too fast)

    source

    • jarfil@beehaw.org ⁨1⁩ ⁨day⁩ ago

      Depending on how much is “too much power”, people might still want to purchase them at a discount for self-hosting purposes. The future is most likely to go through a decentralization of AI services, with higher efficiency large providers, combined with lower efficiency edge nodes for low and less demanding usage.

      source
• stealth_cookies@lemmy.ca ⁨2⁩ ⁨days⁩ ago

  This post assumes way too much and gives the businesses more credit than they deserve. All the AI companies are looking for is maximizing compute per unit of rack space. They are only operating under the goal of winning the AI race to become wildly profitable and powerful. There is no consideration for any bankruptcy proceedings if/when the AI boom comes crashing down as that isn’t their problem at that point.

  The GPUs used for these LLMs are simply not in a form factor that could be used for consumer devices, they fit in racks that use far more power than a home computer would be able to provide.Making them that way would be utterly idiotic.

  source

  • iloveDigit@piefed.social ⁨2⁩ ⁨days⁩ ago

    People years ago said the cards used for crypto mining took too much power and cooling for gamers, but later we got gaming cards that take pretty much the same power and cooling as some of them.

    I hope you’re right and that doesn’t happen this time. It’s gone too far already.

    source
• thatsTheCatch@lemmy.nz ⁨2⁩ ⁨days⁩ ago

  I’m confused — GPUs main function is to be able to do lot’s of calculations in parallel, vs a CPU which does one thing at a time (simplistically).

  GPUs aren’t only used solely for video, it’s just that graphics are an excellent use case for this type of processing.

  So I don’t think AI companies are buying GPUs for video output and more because they can process lots of training calculations in parallel. Like how bitcoin miners use GPUs even though there’s no video involved in that

  source

  • TehPers@beehaw.org ⁨2⁩ ⁨days⁩ ago

    To be more specific here, GPUs are really, really good at linear algebra. They multiply matrices and vectors as single operations. CPUs can often do some SIMD operations, but not nearly as well or as many.

    Video games do a lot of LA in order to render scenes. At the bare minimum, each model vertex is being multiplied by matrices to convert from world space to screen space, clip space, NDC, etc which are calculated based on the properties of your camera and projection type.

    ML also does a lot of LA. Neural nets, for example. are literally a sequence of matrix multiplications. A very simple neural net works by taking a vector representing an input (or matrix for multiple inputs), multiplies that by a matrix representing a node’s weights, then passes the result to an activation function. Then does that a bunch more times.

    Both functions want GPUs, but both need different things from it. AI wants GPUs with huge amounts of memory (for these huge models) which are optimized for data center usage (using cooling designed for racks). Games want GPUs that don’t need to have terabytes of VRAM, but which should be fast at calculating, fast at transferring data between CPU and GPU, and capable of running many shader programs in parallel (so that you can render more pixels at a time, for example).

    source

    • iloveDigit@piefed.social ⁨2⁩ ⁨days⁩ ago

      This doesn’t mean it would be near useless to just add video outputs to neural net cards though.

      Used data center GPUs might be equivalent to a low end or outdated GPU with extra VRAM, but there would be so many of them on the market, you’d see stuff like games being optimized differently to make use of them.

      source

      • -> View More Comments
  • iloveDigit@piefed.social ⁨2⁩ ⁨days⁩ ago

    This all applies to cryptocurrency miners too.

    In fact, it might be even more relevant there, because crypto miners compete so hard on electric bill costs, they definitely have to plan on liquidating equipment when it gets old enough, even if it still works.

    If this was all real business and not just the authorities controlling people, Nvidia would have competition offering similar cards with video outputs for a few cents more because they make more business sense. But instead, it would be super expensive to add video outputs to specialized cards, because it would “cannibalize sales” for graphics cards later.

    source

    • FaceDeer@fedia.io ⁨2⁩ ⁨days⁩ ago

      No major cryptocurrency has used GPUs for mining for many years. Bitcoin uses completely custom ASICs and Ethereum switched away from proof of work entirely.

      source

      • -> View More Comments
• Sxan@piefed.zip ⁨2⁩ ⁨days⁩ ago

  Used they wouldn’t be worth much anyway. At those scales and speeds, GPUs burn themselves up fairly quickly. Average lifespan for a modern gaming GPU is 5-8 years, and that assumes normal use - the GPUs in these AI centers are burning 24/7. I suspect when they get swapped out it’ll be because they’re failing.

  You wouldn’t want a used data center AI GPU, anyway.

  source
• iloveDigit@piefed.social ⁨2⁩ ⁨days⁩ ago

  More detailed explanation of the dynamics behind this:

  Companies like OpenAI would be smart to pay a few extra cents per unit for video outputs to help prevent depreciation if they need to liquidate assets later. But it probably costs way more than a few extra cents per unit, because the chipmakers have a monopoly. The chipmakers know nobody else has been allowed to build the same level of chip fabrication tech they have, so all customers will have to pay the prices they set.

  Therefore, if openAI wanted to add video outputs to their cards, Nvidia has no financial reason to allow prices lower than the revenue that would be lost by consumers not needing to pay for the same chips to be made again with video outputs later.

  But that’s not some genius calculation Nvidia is making to outsmart everyone. You don’t have to be smart or superhumanly profit-maximizing to figure out that you’re being given the authority to set prices when it is given to you. Furthermore, a random group of mid-IQ psychopaths would never be given authority to set these prices in a market of real businesses and genius entrepreneurs competing to maximize profit.

  You’re supposed to believe:

  1. The authorities printed this money because they wanted more money
  2. They forgot the goal was to have more money for themselves, and they accidentally distributed the money through things like wages and bank loans while running out of ink
  3. For the millionth time, they remembered they want as much money as possible and they need all that money back because ink will never be sold on the open market again
  4. They don’t want to just force you to give all the money back at gunpoint, that would risk destabilizing their authority, so they have to basically outsmart you into giving the money back, just like all the other times this has happened
  5. Every time this happened in the past, they split up into corporate teams to make it a game seeing who could come up with the smartest plans to get money from you, and that’s become a strong standing tradition where these corporate teams stay in place eager to play more money game
  6. They make today’s money the “ball” for yet another session of the money game
  7. The authorities that went in Team Nvidia for this round are winning this round because they figured out a genius strategy to scoop up a bunch of money with this AI bubble, and then years later, also finish getting the last bits of money from people buying GPUs again because the AI bubble cards don’t have video outputs
  8. Remember, they are not doing this on purpose, these are just the kind of mistakes you get out of geniuses who are obsessed with getting as much money as possible. They don’t mean to forget the goal of printing the money and accidentally distribute it through wages and bank loans and stuff, but even the smartest most money-obsessed person just forgets stuff like that over and over. They don’t do this for the love of the game, they do it because they’re very smart and they want more money, but just not smart enough or desperate enough for more money to avoid forgetting why they printed it and accidentally distributing it yet again, triggering yet another round of this money game that they insist is a side effect of the real goal of more money. They’re also not smart enough to realize affordable ink sources will probably be found someday to print more money again, just like all the other times this has happened.

  What actually happened:

  1. The authorities printed more money because they wanted stuff people would give them for it
  2. People will give them stuff for the money mainly because they want the stuff they can get with it too, but they can’t just print it like the authorities can, so they get the money by giving the authorities stuff so they can use the money from the authorities to get other stuff
  3. Therefore, if the authorities want to keep offering money in return for stuff, they should also offer stuff in return for money - and they can make it profitable for themselves by offering as little money as possible in return for stuff, and as little stuff as possible in return for money
  4. They also have authority over the production of computer equipment
  5. They wanted a bunch of computer equipment for themselves, but not for you, so they made it special for themselves and not for you
  6. This way they will still be able to offer as little as possible for the most money possible
  7. They didn’t distribute the money accidentally after forgetting the goal was to have as much money as possible - they printed the money to offer it for other stuff they wanted. They didn’t and still don’t have some strange desire for more of the money they can freely print, and if they did, it would be weird that they’re trying to get it from people they gave it to instead of just printing more, since they’re well aware ink is available and affordable to just print more

  source