NVIDIA hasn’t actually announced a name for its next-generation GeForces, but they’re likely to be known as the GeForce RTX 40 series. There’s been a ton of buzz about the top-end of that family. The fastest card is expected to be called the “GeForce RTX 4090 Ti,” and as we’ve reported, it could draw as much as 800 watts.
The next-generation parts from the green team will be using its Ada graphics architecture, named after Ada Lovelace. Everyone expects the high-end SKUs to launch first, so that’s what much of the news has focused on, but most people probably aren’t going to be buying an RTX 4090. That makes us much more interested in rumors and leaks about the lower-end cards, particularly the x70 or x70 Ti models.
Well, that’s just what we have today. Courtesy once more of loose-lipped leaker kopite7kimi, these are the first rumors dedicated to what is most likely to become the GeForce RTX 4070 Ti—assuming the leak is legitimate. You can see his tweet below:
To break it down for you, AD104 refers to the third-largest Ada Lovelace GPU. Earlier leaks have implied that AD103 will be used for the x80 series parts, while AD102, the largest chip, will be used primarily in x90 parts. This tweet claims there will be a graphics card carrying an AD104 processors with all sixty shader modules active, giving it 7,680 FP32s.
The leaker claims that the card will come with 12 GB of GDDR6X memory running at 21 Gbps per-pin, giving it memory bandwidth of 504 GB/second. While, incredibly, that actually sounds kind of moderate these days, the increase in L2 cache from 4MB of GA104 to 48 MB on AD104 probably will help make up some of the difference.
The most concerning fact in this leak is that the AD104 GPU could use as much as 400 watts of power. It’s important to keep these numbers in context, though. All of the biggest, thirstiest GPUs before the GeForce RTX 4090 Ti were well under 400 watts. That includes the GeForce GTX 580, the Radeon R9 390X, every Titan card, and even the Radeon Vega Frontier Edition. To break 400 watts you had to slap a pair of GPUs on a single board, like the Radeon Pro Duo.
This, meanwhile, is a single upper-midrange GPU that could draw as much as 400 watts. That’s particularly concerning in light of the fact that, as far as we know, NVIDIA’s Ada GPUs are fabricated on TSMC’s bleeding-edge 4N process. That’s a refined version of the 5N process that AMD is using for its Navi 3x parts, and that makes it fully two nodes ahead of the refined-10nm Samsung 8N process that NVIDIA used for Ampere.
We’ve heard that NVIDIA might be pushing the top-end Ada parts to sky-high clock rates circa 3 GHz. It’s possible that the company is doing the same thing across its product portfolio. While AD104 is a smaller chip than GA102, the excess clock rate could certainly make up the performance to have it meet the leaker’s expectations.