With Nvidia’s recent RTX 3050 likely the last in the Ampere family, eyes have turned to what lies ahead with RTX 4000. Unfortunately, that glimmer on the horizon might be a giant fireball, possibly emblazoned with the words “Ada Lovelace.” Nvidia’s next-generation RTX 4000 GPU architecture is reportedly so power-hungry we hope the rumors aren’t true. If they are, an awful lot of gamers may need power supply upgrades just to run the card.
According to noted Twitter GPU soothsayer Greymon55 via Videocardz, Nvidia’s upcoming AD102 chip, which represents the full version of it’s next-gen architecture, will come in three versions. The variations will reportedly consume 450W, 650W, and an eye-popping 850W, though Greymon55 adds that this is subject to change. They also noted in a later tweet that what they are posting aren’t always “leaks,” but more just their personal “thoughts and guesses,” so go and find the biggest grain of salt you can to consume with this information.
Interestingly, another famous GPU-predicting Twitter account chimed in as well, saying they’ve heard similar rumors but with slightly lower TGP values. Twitter user @kopite7kimi noted, “That’s just a rumor. I’ve heard 450/600/800W for 80/80Ti/90 before. But everything is not confirmed.” We should pause here and appreciate the irony of one rumor monger telling another it’s “just a rumor.” Also, if you’re thinking it would be just Nvidia launching a GPU with these types of insane power requirements, think again. Greymon55 notes that they would only do it to match or best AMD, and wouldn’t do it if AMD planned to launch 350W RDNA3 cards, for example, which are due around the same time as Nvidia’s Lovelace cards. They wrote on Twitter that for AMD’s next-gen “power consumption will not be low, but probably not crazy.” Going back to the Nvidia rumors, Greymon55 also tweeted that anyone who expected to buy an AD1020-based GPU, such as an RTX 4080 Ti or 4090, would need a 1,500W power supply.
As you may recall, we previously reported on these two Twitter accounts previously posted rumored specs for both Nvidia’s AD102 and AMD’s RDNA3 cards, and if they are anywhere close to the truth both camps are taking the “all guns blazing” route for their upcoming GPUs. What will be particularly interesting in the next round of the GPU wars is both companies are using TSMC’s 5nm process, which is the first time in recent memory that both companies will share manufacturing technology for competing product lines. We wrote at the time that these GPUs could end up consuming around 500W, which seemed high at the time. Now there are rumors that the incoming RTX 3090 Ti might come close to that number itself, with rumors circulating it might consume 450W. It makes sense that power consumption might increase in Nvidia’s next-gen products despite moving to a smaller manufacturing node (8nm to 5nm). The death of Dennard scaling in 2004 has left GPU manufacturers unable to reduce voltage to compensate for higher clock speeds and we can see that reflected in the way GPU TDPs have steadily climbed over the last few years.