Nvidia could have an RTX Titan graphics card inbound with its next-gen Blackwell line-up, if a new rumor is right.
This comes from RedGamingTech (RGT) on YouTube, who claims an RTX Titan AI is in the works, and that contention is backed up by another well-known leaker, Kopite7kimi (on X, as you can see below).
RGT also brings us performance uplift estimates for the RTX 5000 range, and according to the YouTuber, the RTX Titan AI for Blackwell will be 63% faster than the current flagship GPU, the RTX 4090. (Keep a keen sense of skepticism around all this info, by the way – including the potential existence of a Titan – and as the leaker observes, Nvidia’s next-gen graphics card configurations could still be altered before release, even in this relatively late stage in the development process).
RGT also tells us that the RTX 5090 will be a 48% uplift over the 4090, and the RTX 5080 will be around 29% quicker than the RTX 4080 Super. Similarly, the RTX 5070 will apparently be 26% faster compared to the RTX 4070 Super.
The approximate 50% performance uplift for the RTX 5090 is notably less than some other rumors – which have suggested an up to 70% performance boost – but most speculation has landed in the 50% to 70% area, so it is in line with what we’ve previously been primed to expect.
What we don’t know, as the leaker admits, is whether this is based on just one benchmark or some kind of average – the figures were just presented to RGT as is. Therefore, if it is just a single benchmark (maybe a synthetic one), we need to add even more seasoning.
What is made clear is that these are estimates for rasterization performance, meaning ‘normal’ rendering as opposed to ray tracing. In the case of the latter, Nvidia’s RTX 5000 graphics cards will make bigger leaps in performance (which we’d expect, in fairness – ray tracing is very much Team Green’s strength, and an area it’s keen to push hard with).
Analysis: Titanic ambitions?
If the RTX Titan AI is indeed real, it’ll use the same GB202 chip as the RTX 5090, but the Titan will have the full amount of CUDA Cores enabled (whereas as we’ve already heard via the rumor mill, the 5090 is set to cut down the full count a bit). Odds are the Titan would also have a wider memory bus (512-bit, rather than the rumored 448-bit in the RTX 5090) and more VRAM (maybe a whole lot more). As RGT contends, this will make it around 10% faster than the RTX 5090.
Although you can bet it’ll be more than 10% more expensive than the 5090, guaranteed – this will be one eye-wateringly priced graphics card, if it emerges. Indeed, the RTX 5090 is already rumored to be notching up the cost of a flagship Nvidia GPU from the level the 4090 reached.
Of course, if the idea of an RTX Titan AI has you excited, remember – it’s not a gaming graphics card, or that’s not what it’s built for anyway, as the ‘AI’ name makes clear. While it might be the fastest consumer desktop GPU, it’s aimed at professional usage, and will be loaded up with VRAM to suit the likes of creatives, and more heavyweight tasks like rendering.
However, while it isn’t a GeForce gaming GPU by design, that won’t stop PC gamers picking it up for bragging rights as the fastest desktop graphics card out there – even if the outlay for a bit of extra gaming pep won’t make any sense at all, value proposition-wise.
The RTX Titan AI for Blackwell may not happen anyway – even if the leak is real, this could be Nvidia toying with the concept, as happened with the current Lovelace generation. If you recall, an RTX 4090 Ti (or Super) and new Titan were rumored substantially, and as VideoCardz (which spotted the above video and tweet) pointed out, there were even prototype designs for the latter.
In short, we remain pretty skeptical, if only because we’re not sure Nvidia will find space in its next-gen GPU line-up for such a Titan, when it likely already has plenty on the menu for powerful AI graphics cards based on Blackwell – and there’s only so much silicon to go round.
You might also like
https://cdn.mos.cms.futurecdn.net/2MVE55kX5okLKHbpxjbN3c-1200-80.jpg
Source link
