Regale mit KI-Chips sind verdammt schwer | Alte Rechenzentren können physisch keine Reihen von GPUs unterstützen, was einer der Gründe für den massiven Ausbau von KI-Rechenzentren ist.
https://www.theverge.com/ai-artificial-intelligence/844966/heavy-ai-data-center-buildout
8 Kommentare
>Chris Brown, chief technical officer at Uptime Institute, summarized the situation: “We can retrofit the old ones to an extent, but not to the extent that a lot of these AI factories need.” Small sections of small data centers can accommodate small AI-focused workloads for a single Fortune 500 company, for example, he said. “But most of the time what it’s going to mean is bulldozing the building and starting over from scratch,” Brown said.
So that seems a bit extreme. If they are heavy why bulldoze the building and not just build new rack infrastructure.
Translation: new one has to be online before the old one is decommissioned.
The other issue is energy density. What was below 100 kW per rack is now approaching 200 kW per rack and on its way to 1MW per rack in five or six years. Everything from power to telecom to cooling needs to expand.
Just wild what our industry is willing to do for “AI”.
Complete BS. Power and cooling problematic: sure. Elissa@theverge got trolled.
Throwing all this money and energy at various companies that’ll have to purchase components over and over again because of increased demand only to get a shitty chat bot that tells you it can’t understand you.
Weight? Seriously? I can see power but I used to work storage racks full of legacy spindles where each 2u was about 70 pounds racked on raised floors. GPU’s are heavier than that?
If we only had the technology to build sturdier racks…