Recent generations of flagship graphics cards have seen an upward trend in maximum power consumption: the RTX 4090 is over 450W, the RTX 3090 Ti is similar, and AMD's RX 7900 XTX is at 355W. Factory overclocked models are usually slightly higher than that. But these are rookie numbers; Nvidia's next-generation AI accelerator GB200 can reportedly draw up to 1,000W.

The source for the 1,000W claim is a statement made by Dell COO Jeffrey Clark at Dell's Q4 2024 earnings call (via The Register). This 1,000W claim is not just a rumor, as COO Clark knows more than anyone else.

Clark's actual statement was, "Our hallmark on the thermal side is that we don't need direct liquid cooling to achieve an energy density of 1,000 watts per GPU. That will be achieved next year with the B200."

Note that although he mentions the B200, the B200 is actually not on Nvidia's roadmap; Nvidia's next-generation enterprise flagship GPU is called the B100. It is based on the Blackwell architecture. However, there is a GB200 that combines a B100 GPU with a Grace CPU. Clark simply misspoke and most likely refers to the latter.

GPUs for data centers are very different from GPUs for consumer gaming, and things like fan speed, physical size, and power consumption are secondary to performance. the GB200 is all about performance. the Blackwell-based The different variants of accelerators will ultimately form the backbone of cloud AI systems in 2025 and beyond.

Everyone now knows how big business AI is: Nvidia is now the world's third largest company by market capitalization, recently overtaking Saudi Arabia-based oil giant Aramco, with only Apple and Microsoft ahead of it. And it is catching up fast, as the demand for high-performance AI accelerators remains insatiable.

It is unlikely that 1,000W desktop graphics cards will ever emerge. For the average ATX PC, 1,000W is too far away, but with 450W becoming the new baseline for high-end gaming GPUs, expect it to step up again; if Nvidia were to release the RTX 4090 Ti, 600W or more could be a reality We can also imagine that See the leaked image of the huge quad-slot Founders Edition cooler.

Personally, I hate the idea of a card using that much power. I regularly use a 450W RTX 4090 and it gets really hot in my office room during prolonged use, I don't want to think about what a 600W+ card would do during the warmer months of the year, I'm not sure how much power it would use, but I'm sure it would be a lot.

Whatever form the RTX 5090 ultimately takes, it will be a hot, power-hungry card. With TDPs exceeding 500W for high-end Blackwell-based GPUs, quad-slot or AIO-cooled cards are likely to dominate, especially overclocked models.

Hopefully cards like the RTX 5060 and RTX 5070 will maintain the excellent power efficiency and performance-per-watt characteristics of cards like the RTX 4060 and the recently released RTX 4070 Super.