NVIDIA's Blackwell graphics cards, expected to be GeForce RTX 5000 models, could be over twice as quick as current-gen Lovelace models - and we've heard a pile of other rumors about Team Green's next-gen GPUs to boot.
This comes courtesy of Wccftech, with the tech site keeping its ear to the ground and flagging up a bunch of info - approach all this very cautiously, naturally - from Twitter and two prominent YouTube leakers, Moore's Law is Dead (MLID) and RedGamingTech (RGT).
Bearing firmly in mind that it's still early stages for Blackwell, we hear from MLID that NVIDIA is likely to go with a monolithic GPU (single chip) for the next-gen flagship (RTX 5090). That idea has been floated in the past via the rumor mill, notably from renowned Twitter leaker Kopite7kimi, but there's a twist here - MLID doesn't completely rule out chiplets.
MLID tells us there's a "remote chance" that the RTX 5090 may follow the route AMD took with its RX 7900 graphics cards and use chiplets, but this is unlikely. Furthermore, even if the RTX 5090 eventually turns out to be a chiplet design, we're assured that other Blackwell GPUs won't be.
RGT's video puts forward some more eyebrow-raising theories, including a big claim on the performance front. Namely that Blackwell GPUs could hit 3GHz clock speeds with some cards, and be twice as fast for frame rates as current-gen Lovelace, or maybe even slightly quicker than that.
Did we say big claim? Strike that; let's call it a massive one. That said, RGT does heavily caveat this, saying that despite hearing this on the grapevine, it's still very early days, and it's unclear what kind of performance this guestimate pertains to. Is it ray tracing performance, for example? We simply don't know and wouldn't put much stock in any exact figures at this point anyway.
This is, however, an exciting hint that NVIDIA has something big in the works for RTX 5000. RGT reckons that Team Green's next-gen graphics cards will come with a major overhaul to the architecture - and drive even harder for ray tracing performance - while being built on TSMC's 3nm process.
On that last point, Kopite7kimi notably disagrees and recently tweeted that Blackwell won't use 3nm. This underlines the early stage we're at with these rumors contradicting each other, and that we should be very careful about taking any of this speculation too seriously. At least for the time being, anyway.
Indeed, Kopite7kimi recently tweeted that it's too early to discuss Blackwell, and we can agree on that broadly. But at the same time, there's no harm in putting out some feelers to get an idea of where NVIDIA might be headed at this stage.
Suffice it to say that the early signs are positive, but really, as we look towards the RTX 5000 range, we're hoping for one thing in the main - prices that aren't majorly stacked against gamers.
Not to mention better budget graphics cards in general, but that area of the market is one that NVIDIA doesn't seem to be bothered about of late. Although you never know - we're still holding out some hope for the RTX 4060 to be relatively affordable (the RTX 4070, on the other hand, would appear to be a very different story).