, the rumored range to succeed current Lovelace GeForce GPUs, due to arrive in 2025 theoretically as RTX 5000 graphics cards, might be missing a model.
spotted a rumor on the Chiphell forums, a Chinese source that needs large amounts of seasoning – and we mean hefty, but in this case, the speculation is backed up by a regular leaker on X (formerly known as Twitter), Kopite7kimi.
The contention, as aired by Kopite7kimi and ‘Panzerlied’ on Chiphell, is that there will be no x04 model with Blackwell – but an x05 product instead. What this means is that Nvidia won’t make an equivalent of the AD104 chip (which sat inside the RTX 4070), but instead, is going to produce a GB205 (GB denotes Blackwell) instead.
Kopite7kimi tells us that the chips that Blackwell graphics cards will be built with are as follows: GB202, GB203, GB205, GB206, GB207.
That mirrors Lovelace and Ampere, except for the x04 being switched to x05 (and the change to 2xx, which we’ll come back to later).
Analysis: What’s in a name?
So, what does this actually mean? Well, as mentioned AD104 is the third most powerful GPU for Lovelace, driving the RTX 4070 (and sitting underneath AD102, which is the RTX 4090’s chip, and AD103, which is in the RTX 4080 – though the 12GB spin on the latter actually uses AD104, and we often see nuances like this).
This doesn’t mean there won’t be an RTX 5070 graphics card though. For starters, we should note that it doesn’t necessarily mean anything – it might just be a whole lot of nonsense chatter from the grapevine that disappears into the black hole of ‘erroneous early rumors’ in the future.
But, assuming Nvidia will make a GB205 chip rather than a GB204, could this indicate that while there might be an RTX 5070 – let’s face it, Nvidia is very unlikely to scrap one of its stalwart GPU models, currently top of our best graphics cards ranking – it may be slightly weaker in comparison to what we traditionally expect?
That’d appear to be a simple conclusion to draw here, but it is just that – too simplistic. We simply don’t know how Nvidia is planning to pitch the relative performance levels of the Blackwell range yet (it’s entirely possible Team Green still isn’t sure on this itself, at this point).
So, for example, an RTX 5070 model could use the GB203 chip as well as the RTX 5080, perhaps. Okay, so that’s unlikely, yes, but the point is that it isn’t totally unfeasible. Or, Blackwell could be such a leap forward architecturally that the GB205 chip could give an RTX 5070 every bit of the performance gain it requires gen-on-gen compared to Lovelace and the RTX 4070.
The worry is that this apparently represents a “new market positioning” for Nvidia, the Chiphell leaker informs us, and there is a hint that if Team Green is switching to a lower number with this third-best chip, then this is a suggestion that the mid-range (well, upper-mid-range) of Blackwell could be pitched a little more modestly performance-wise.
What’s also a bit strange is that the chip names for Blackwell here are GB2xx, when they would normally be GB1xx (as in AD104). Why the uptick with that initial digit? Apparently it’s because Blackwell will be both pro and consumer GeForce GPUs, rather than splitting the designation (as with Hopper and Lovelace). So GB1xx models will be data center and heavyweight usage, and GB2xx models will be gaming GPUs, or that’s the theory anyway.
Given how far out Blackwell still is, at this point – as mentioned, Nvidia probably hasn’t firmed up a lot of the details itself yet – we’d take all this with more skepticism than normal for a GPU rumor.
What does give us a further pause for thought here is that if Nvidia is somehow looking to rein in this mid-range area, this is where AMD may come in with its toughest competition with RDNA 4 – as Team Red purportedly won’t have a high-end Navi 41 offering. In other words, this is where AMD’s best efforts will be going with next-gen GPUs, so wouldn’t Nvidia be wise to, if anything, drive harder here?
Alternatively, the other thread of an argument we’ve been seeing of late is that consumer GPUs are becoming much less relevant to Nvidia (and AMD, to some extent) anyway, given the far greater profit levels to be made with graphics cards on the AI side. Arguably, this is why RTX 4000 GPUs have been priced so highly – Nvidia might be losing goodwill from gamers, but, well, the truth is that it simply doesn’t care all that much.
Via PC Gamer