Fresh speculation about the next-generation of Nvidia graphics cards has popped up over the weekend, with new rumours of a GTX 1170 doing the rounds. I mean, of course it has, that’s the PC gaming tech world’s favourite pastime right now: the life and times of graphics cards. Which GPUs are available, how much they cost, who they’re sleeping with, what will the next-gen look like, and when are they going to bare all in a tabloid exposé?
If you need a new GPU right now, pick one of the best graphics cards available today.
The new GTX 1170 specs have come from the ever-reliable WCCFTech, and seems to be formed from as much speculation and guesswork as the GTX 1180 ‘leak’ that seemed to get everyone all excited last week. What the WCCFTech piece is essentially saying is that the GTX 1170 is going to be... a graphics card. And they then typed out some numbers plucked from the air above their keyboard.
They don’t seem to have any real concrete information, but are still betting on the Nvidia Turing rumour delivering a 12nm GT104 GPU that measures around 400mm2 and packs in 2,688 CUDA cores.
It’ll also have somewhere between 8GB and 16GB of 12Gbps GDDR6 memory (really, at those prices?!), some sort of loose guess at a vague base frequency, which has basically cribbed from the current Pascal cards, and the same for boost clock speed. Oh, and a TDP of between 140 and 160W. My, there’s a lot of margin for error built into these leaks…
WCCFTech, of course, can attribute no source to these numbers, just that they were provided specifically to them. Someone as irked by this baseless speculation as I is a new poster on the Overclockers UK forum with the name Nv Gfx Prgmr (seems 100% legit).
They rail against such rumour mill sites, and say that the supposedly leaked specs are completely false. Good for him.
He does then go and ruin it all with his own baseless speculation, claiming to know that there will be no new 12nm gaming cards released this year and that “Nvidia's new gaming GPUs will be on 7nm and their current release window is set for late November.”
As much as that is either blatant under-the-bridge trollism, or rumour-mongering of the highest order, it’s still not necessarily beyond the realms of possibility for Nvidia to skip 12nm in favour of 7nm for their next generation of gaming graphics cards. Especially as, by the end of the year, they will have essentially skipped an entire generation of gaming GPUs by holding onto Pascal for so very long.
It might seem implausible, but hear me out.
TSMC, the likely manufacturers of Nvidia’s graphics silicon, have already started volume production of 7nm processors, as C. C. Wei, TSMC President told financial analysts last month, “More than 50 products tape-outs has [sic] been planned by end of this year,” he says, “from applications across mobile, server CPU, network processor, gaming, GPU, PGA, cryptocurrency, automotive and AI. Our 7nm is already in volume production.”
Now, we know that AMD have already said they’re using TSMC’s 7nm design for their upcoming product lines, splitting them up with GlobalFoundries, and it seems likely that their 7nm Vega GPUs will be coming from TSMC, with this year’s 7nm Zen 2 samples expected to be built by GloFo.
So 7nm is already very much ‘a thing’ and with so much time spent at the forefront of graphics technology Nvidia aren’t going to want to sit too far behind the competition. To be fair, AMD have already said their 7nm GPU isn’t going to be turned into a high-volume gaming card, and potentially never will, but that is more because of its inextricable ties to the ultra-expensive HBM2. As young, and as expensive, as GDDR6 is as an emerging graphics memory tech it’s unlikely to have quite the prohibitive costs of HBM2.
It's not going to be the super-exciting EUV-toting 7nm silicon that Intel, Samsung, GlobalFoundries, and TSMC are introducing in the next few years, but it'll still offer a 60% drop in power consumption, and a potential 70% drop in die size.
But if AMD are showing off any 7nm graphics silicon this year, for machine learning or otherwise, then Jen-Hsun might not be too happy releasing a brand new gaming range on a lithography that’s very obviously last-gen by that point.
Hubris can be a dangerous thing, however, as can pairing a new graphics architecture with a brand new production lithography - just look at the hot mess that was the GTX 480. And Nvidia are unlikely to want to lead with a new architecture on a new process node with a gaming GPU instead of a lower-volume, professional-class graphics card.
However, if the next generation of GeForce GPU is heavily based on the current Nvidia Volta architecture - as is widely expected, whatever name it gets given - then it won’t really be a new design. By that time Volta will be a far more mature GPU design, and the engineers could well have a bead on just how well it could translate down to a smaller, high-volume 7nm node.
As much as it’s all baseless speculation right now (sorry, Nv Gfx Prgmr), skipping a short-lived 12nm GPU, in favour of a 7nm design that could last another two years or more, does make a lot of sense. Will they do it? Right now, who knows. Realistically, it's unlikely, the timing is possibly too tight, and it's probably more of a gamble than Nvidia needs to make right now. But sadly we're probably not going to find out for sure until August/September time at the earliest. Gamescom maybe?