Tuesday, October 17th 2023
NVIDIA Readies GeForce RTX 4070 SUPER, RTX 4070 Ti SUPER, and RTX 4080 SUPER
NVIDIA is rumored to be working on a refresh of the higher end of its GeForce RTX 40-series "Ada" series, according to hongxing2020, a reliable source with NVIDIA leaks. The company could be bringing back the SUPER brand extension that it introduced with the RTX 20-series. As many as three SKUs are on the radar—GeForce RTX 4080 SUPER, GeForce RTX 4070 Ti SUPER, and the GeForce RTX 4070 SUPER.
There is no word on when the company plans to release these, or what their specifications are, but we can certainly speculate. The current RTX 4080, while based on the AD103 silicon, doesn't max it out—it uses 76 out of 80 SM (streaming multiprocessors) available on the silicon, but we doubt if those extra 4 SM could drive up enough performance to make a whole new SKU, especially given that the 256-bit memory bus of the AD103 is maxed out. We predict that the RTX 4080 SUPER could be based on the larger AD102 silicon that physically has 144 SM that the current RTX 4090 uses 128 out of. NVIDIA has the opportunity to pick an SM count such as, say, 96. AD102 also has a wider 384-bit memory bus, giving NVIDIA the option of either giving the RTX 4080 SUPER the same 24 GB memory configuration as the RTX 4090, or even 20 GB, across a 320-bit memory bus.
As for the RTX 4070 Ti SUPER (an oddly named SKU that combines both the Ti and SUPER brand extensions), considering that the current RTX 4070 Ti maxes out the AD104 silicon (60 SM), it stands to reason that this new SKU could be based on AD103, with a lower SM count than the 76 of the RTX 4080. It remains to be seen if NVIDIA utilizes the full 256-bit memory bus. As for the RTX 4070 SUPER, there exists a rather wide gap between the current RTX 4070 Ti and the current RTX 4070. While the RTX 4070 Ti uses all 60 SM on the AD104, the RTX 4070 only uses 46. This gives NVIDIA the opportunity to get closer to the RTX 4070 Ti in SM counts, and bolster against AMD's Radeon RX 7800 XT.
Sources:
hongxing2020 (Twitter), VideoCardz
There is no word on when the company plans to release these, or what their specifications are, but we can certainly speculate. The current RTX 4080, while based on the AD103 silicon, doesn't max it out—it uses 76 out of 80 SM (streaming multiprocessors) available on the silicon, but we doubt if those extra 4 SM could drive up enough performance to make a whole new SKU, especially given that the 256-bit memory bus of the AD103 is maxed out. We predict that the RTX 4080 SUPER could be based on the larger AD102 silicon that physically has 144 SM that the current RTX 4090 uses 128 out of. NVIDIA has the opportunity to pick an SM count such as, say, 96. AD102 also has a wider 384-bit memory bus, giving NVIDIA the option of either giving the RTX 4080 SUPER the same 24 GB memory configuration as the RTX 4090, or even 20 GB, across a 320-bit memory bus.
76 Comments on NVIDIA Readies GeForce RTX 4070 SUPER, RTX 4070 Ti SUPER, and RTX 4080 SUPER
IMO the only one that really makes sense here is 4070 SUPER, again due to yield improvements. If NVIDIA can enable 50+ of 60 SMs on AD104 instead of 46, that would bring 4070S a lot closer to 7800 XT, which is really the only card that NVIDIA struggles against right now in terms of price/perf.
Looks increasingly like the 50x0 series will be the key to good UE5 4K + RT performance.
Maybe the 4070Ti and 4070 Super come with 16GB VRAM. Yeah I know I am just dreaming...although if the 4070Ti Super is made with the AD103 16GB could be true...Not even the 4090 can do that... Remnant2 runs at 35-40FPS and the game doesn't even have RT...
Even the RTX5090 might not be able to do that...unless it has >70% more performance...
If real the easy 4070 Super upgrade would be to use AD104 and cut it less so performance is ahead of the 7800XT in raster and clearly ahead in RT. Sell it for $650 and it would probably be fine although 12GB of VRAM is a bit ugh at that kind of price, especially since FG and RT use up more VRAM.
The 4070 Ti Super would need to be further cut AD103, give it 16GB of VRAM, make it faster than the 7900XT / 3090Ti in 4K and you are sorted. Could probably slot in at $800 and the OG 4070Ti can either be reduced in price or just go EOL. This is probably what the 4070 Ti should have been to begin with IMO.
The 4080 Super should aim to clearly beat the 7900XTX in raster, I expect a cut down AD102 with 20GB of VRAM here and it won't threaten the 4090, gets to keep the current 4080 price point so there is still an upsell to the 4090.
Existing 4080 can fill in at $1,000.
Stack would be like this
4090 - $1,600
4080 Super - $1,200
4080 - $1,000
4070 Ti Super - $850
4070 Ti - $750
4070 Super - $650
4070 - $550
4060 Ti 16GB - $450
4060 Ti 8GB - $375
4060 8GB - $300
4060Ti 16GB is already at that price, the 4070 is already at that price, 4070Ti is approaching that price and the 4080 is clearly overpriced at $1,200 so I think it works if NV have enough wafer capacity to supply their pro clients and expand their dGPU SKUs.
"Moore's Law is dead … It's completely over, and so the idea that a chip is going to go down in cost over time, unfortunately, is a story of the past."
So I could clearly see the existing models that are too close together actually go UP in price to make room for the models below them! There isn't much room between RTX 4080 and RTX 4070 Ti for a RTX 4070 Ti Super in between? Increase the price of 4080! And so on...
I will not waste my money on another RT card that you have to turn the RT off just so the game is playable, if these cards were still £600, i'd be up for it, but not at double that. Remember that by this time next year, UE5 games will be far more prevalent, and crappy performance will be more obvious and less tolerated by more people than today.
I will just hold on for the 60x0 series if the 50x0 can't do it, unless AMD surprises me, which, for obvious reasons, I doubt.
Also... anyone noticed 4090 prices seem to be on the rise? :wtf:
I'd not expect anything better than plus 8 percent bang per buck on average compared to non-Super semi-predecessor. Don't forget that Leather Jacket Guy got a wee greedier than he used to be. Most likely discounted "old" Ada GPUs will be much more feasible, especially outside the States."Just buy a 4090."
It is just like when Crysis released, there wasn't really a card that could run it that fast too.
For a reminder when Crysis came out the fastest card could run it with barely 15fps in FullHD. Even 3-way-SLI with GTX 8800 Ultras was only upto 40FPS. :D
So no, I don't expect a current Gen GPU to run games with a next Gen Engine with maxed settings, high resolution and high FPS.
And remember that UE5 is still in active developement!
I have a 1660 in my second PC. I might want to upgrade to a 4070 at some point, especially when i use the 4070 in my mainsystem when i sell my 4090 before the 5090 launch.
I haven't bought a 4070 yet, because there is only 1 model that fits in my 5L SFF-PC and even that one only with some modding on the shroud.
If a 4070 SUPER releases, I hope for lower 4070 prices or shorter/thinner Super models. :D
They don’t even sweat. They’ll just release a couple more skus so they maintain the interest and the prices high.
If I resisted buying a 4080, now I would go all in for a Ti.
UE5 like UE4 is broken. Only a few titles will tame this engine to run in humane systems.
As I've said before, and while indeed it is speculation wrt trusting there is some validity in the rumor mill, whatever AMD/Intel release next year is almost-certainly going to screw up nVIDIA's whole stack (other than 4090), just as Navi32 did for everything below 4070ti (as it should because nVIDIA's performance below that is pathetic when put in context of pricing/older products with similar performance, especially before 4070 received a price cut).
That said, it shouldn't be difficult as nVIDIA's 4070ti/4080 are still egregiously priced. What better way to non-admit this to than to release new skus, readjusting prices wrt that new competition in the process?
Perhaps this won't be entirely on the money, but I can absolutely imagine a scenario where 7700xt/7800xt are ~$300/400 and 7900xt/7900xtx are ~$650/800, while higher-clocked 8192sp parts from AMD/Intel land in the middle. Explain to me how nVIDIA has any hope (beyond short-term thinking RT evangelists) in anything resembling that scenario without dropping prices and releasing new SKUs.
If there truly is a 4070ti ti ti ti super, that will without a doubt be the most-humorously named panic SKU ever created. What do they do, release a 256-bit 72-74SM AD103 just to keep 4080 at a higher margin?
I just cannot imagine a scenario where nVIDIA does not get burned for not having a 256-bit/16GB GPU for the ~$400-800 market, which they absolutely should. $800 for 12GB is close to robbery. $1100 for 16GB is lunacy. At some point it's really going to suck for people that bought into AD104. I feel less bad for 4080 owners, as the performance should stay relevant a while even if the price is/was too damn high.
What's also amusing, or perhaps telling, is this trend supposedly continues for Blackwell (there is no GB204). While there is no guarantee that means 128-bit, 192-bit, 384-bit, and 512-bit chips, it could in-fact mean that thing. To me that signifies that nVIDIA knows they're going to be filling that gap with cheap(er) AD103s (and conceivably even AD102s) for quite some time, propping GB102/103 up on a high margin just like they did AD102/103. They really just refuse to release a value product unless they absolutely have to do it, which they won't until AMD/Intel release something competitive in each higher-end market segment. Thankfully, I think Navi4x/BM will probably scratch the itch for many people (1440p high-end settings) until the next console gen at a reasonable console-like price bracket. At least, that's the hope.
Gaming PC market outside the US is bigger than the market inside the US for obvious reasons. And in some countries like Canada and Brazil AMD products are ill-priced. And Intel products higher than A770 Premium don't exist (and this is not even a true 1440p GPU), which leads to successful trades overall. Yes, people buy Ada GPUs in much lesser amounts than they bought predecessors but we currently have no COVID or mining inflated demand and we got a truckload of cheap used GPUs per every suburb so low demand on BNIB GPUs is expected.
All this means you can stoke nGreedia all you please but they don't have any apparent reason to cut their pricing to what seemed normal in 2010s. They won't sell much regardless of pricing due to the fact 2022 to 2024 buyers are way less interested in bleeding edge than before so why not increasing their margins so it will compensate for worsened sells.My bet Blackwell will be the worst generation of the first half of XXI century, with RTX 6000 series onwards being better overall. I might be completely wrong but the feeling we have not yet reached the very bottom but we're one leap close to it doesn't leave me. So RTX "5080" (~40% of full Blackwell potential with 40% being a direct indicator of a *60/Super class GPU) for $1400 give or take 100 bucks is to be expected (RTX "4080" is ~55% of full Ada potential with 55% being a direct indicator of a *70 Ti class GPU). What will they put between 5080 and 5090? A couple GPUs, namely 5080 Super (~50% potential), 5080 Ti (~60% potential) and probably ~80% potential 5090 24 GB (with "normal" 5090 being a 32 GB SKU) or something along those lines. I don't expect them to produce anything below RTX "5060." And RTX "5060" will most probably fail to outperform 4060 Ti.
4080 super 10240/112Rop 16G ~7900xtx
4070 ti super 7680/80Rop 16G ~7900xt
4070 super 6400/72Rop 12G ~7800 xt
and there are some indications that 4070 Ti super can have up 8704 the same as 3080. It's a mess. 4070 16 is king of unlikely.
But this is all academic.
If these rumors are true, Nvidia will slot a new SUPER SKU somewhere in the existing product stack and set a price in between the existing products. The value proposition will be revealed once the new SKUs are benchmarked.
But no one should expect some sort of superior bargain.
For the next couple of years, Nvidia will use every single wafer they can get their hands on and sell every single GPU they produce. Since gross margins are higher for AI accelerators today, they will presumably allocate more resources to those chips. Less consumer graphics card supply will mean higher prices (assuming demand stays the same).
AMD has basically conceded the high end. Nvidia charges more for the RTX 4080 than AMD charges for RX 7900 yet AD103 is a smaller die than Navi 31. If yields are comparable, then Nvidia has higher GM (for the GPU alone) from what I can see.
The real question would be: will they launch the "Super" editions before or after the Christmas season? I guess after, so they can farm the Christmas dollarinos with the old gen.
Naturally Nvidia is under no obligation to ship all of them at the same time. My assumption is that the launch timetable will be influenced by channel inventory of existing products.