• Realitätsverlust@lemmy.zip
    link
    fedilink
    English
    arrow-up
    1
    ·
    10 hours ago

    All that being said, you and I are thinking on a pretty small scale. We’re thinking of liquid cooling on a home PC. For that large a scale, I’m sure the cooling system is pumping from a large reservoir to consistently keep that large a system cool.

    Uuuuuh, no. I’m thinking of datacenters - I am working in one. All of the racks are having their own loops, you don’t want the entire datacenter to power down because the cooling system fails. While it’s not as simple as having an AIO in a consumer PC, it’s still a closed loop, albeit a bigger one. However, I don’t know anything about training models - so maybe there is such a system. I don’t know. It sounds weird to me tho because it’s kinda wasteful without providing any real benefit, and while companies don’t give a shit about the environment, they do give a shit about saving as much money as possible. Having a wasteful cooling system when there are non-wasteful ones seems unnecessary.

    But then again, I don’t know either. I just don’t believe that this study was entirely truthful and tries to make AI look worse than it is. It’s still bad, don’t get me wrong, but this reminds me a lot of PETA studies that sad “ONE BEEF PATTY NEEDS 250.000 LITERS OF WATER!!!” and it turns out they calculated that with a single cow on a massively oversized area and they also included the entire rain falling on that area as “wasted water”.