Nvidia is diving right into the third circle with its gluttonous AI superchips. The new Nvidia B200 AI accelerator is the most power-hungry chip we’ve seen yet with a pull of 1,000W, enough to make a small power grid sweat with fear.
While new GPUs like the RTX 40 Series cards can suck up power like no other, Nvidia’s B200 AI superchips manage to put even the most powerful commercial PC components to shame. It’s time to invest in a gas-fed generator if you wish to power one of these bad boys.
Revealed by Intel Dell COO Jeffrey Clarke, via The Register, Nvidia’s latest chips are expected to pull an insane 1,000W during use. Clarke also revealed that the chip doesn’t seem to require any form of specialist thermal management to hit the large power draw.
“Our characterization in the thermal side, you really don't need direct liquid cooling to get to the energy density of 1,000 watts per GPU. That happens next year with the B200,” the Dell COO told investors.
Obviously, AI chips have a much higher draw call than consumer-grade chips like an Intel i9. Designed to be housed in huge datacenters, these chips have much higher demands than your home PC - and they can get extremely hot.
With generative AI tools such as ChatGPT and Midjourney using as much power as a commercial jet to train, it’s entirely unsurprising just how much juice it takes to fuel Nvidia’s latest chips. As AI gets more advanced, it’s only expected to suck up even more energy.
It’s worth noting that Nvidia has yet to officially reveal its B200 superchips, although it’s almost definitely coming. At the time of writing, Nvidia has released its B100 chip and the GB200 which combines the B100 with a powerful Grace CPU.
Nevertheless, Nvidia is going hard on AI, from its RTX AI features to its Omniverse Metaverse plans. If any company was to make a 1,000W AI chip, it would be this one.