You know what they say, you wait ages for a bus and then two come at once. Nvidia has announced not one but two RTX 4080 graphics cards: the RTX 4080 12GB and RTX 4080 16GB.
While these may sound like two very similar graphics cards with only moderately different memory specifications, when you really get down to the silicon, you might actually consider them to be entirely different graphics cards altogether. That's because the 16GB model comes with many more CUDA Cores than the 12GB model—9,728 to 7,680, respectively—and is priced accordingly.
The cheaper of the two, the 12GB model, will start out at $899. That's $200 more than the original RTX 3080 10GB, though Nvidia is reporting performance around or even better than that of an RTX 3090 Ti, its best GeForce GPU today, at 40 TFLOPs.
The more expensive model is the RTX 4080 16GB, which starts out at $1,199. It comes with 9,728 CUDA Cores, a near-27% increase over the 12GB model. This, Nvidia says, is enough to double the performance of the RTX 3080 Ti for 10% less power. A mighty claim, though at 49TFLOPs the RTX 4080 16GB is nowhere close to actually doubling the 34.1TFLOPs that the RTX 3080 Ti can muster.
There's more to gaming performance than raw TFLOPs, of course, but in this case, Nvidia is citing frame rate data from Microsoft Flight Simulator, Warhammer 40,000 Darktide, and Cyberpunk 2077 with a new ray tracing mode enabled while also using its new DLSS 3 upscaling technology (opens in new tab).
This will help out the RTX 40-series specifically, as these cards can take advantage of a new technology called DLSS Frame Generation to actually create new frames, rather than infer and add information to an already rendered frame.
So we'll have to get these cards in the PC Gamer labs to see what sort of performance they can really muster head-to-head, but we should expect a big leap in rasterised and ray tracing performance—Nvidia has made big improvements to its RT Core (opens in new tab) with the RTX 40-series and how it schedules work on Ada Lovelace GPUs for optimal performance.
Both the RTX 4080 16GB and RTX 4080 12GB will be available from November, and from the pictures we know to expect Founders Edition shrouds available. The RTX 4080 release date is roughly a month on from the release of the even larger RTX 4090 (opens in new tab). That comes with 16,384 CUDA Cores and 24GB of GDDR6X memory, but will set you back $1,599.
And that's not even the largest GPU that Nvidia can reportedly muster. While nothing announced uses this loadout yet, CEO Huang mentioned over 18,000 CUDA Cores were possible from Ada Lovelace on TSMC's 4N process node. That's one massive GPU, but I wonder if we'll ever see it in the GeForce lineup.
But where does this leave the RTX 4070? Where we'd usually see that tier of card, we're instead seeing two models of RTX 4080 for a higher price than the original RTX 3080. So I do wonder when we'll see a cheaper option for the RTX 40-series, and what shape it'll take when we do. So far Nvidia appears to be keeping the RTX 30-series around (opens in new tab) to plug that gap.