NVIDIA Reportedly Working on GeForce RTX 5080 SUPER 24 GB & RTX 5070 SUPER 18 GB Designs

[T0@st]

Mere months after the launch of GeForce RTX 5080 16 GB and RTX 5070 12 GB graphics cards, industry rumors are already swirling about a potential "SUPER" semi-successor. Sections of NVIDIA's previous-gen "Ada Lovelace" GeForce RTX 40-series family received mid-generation refresh treatment around early 2024. It is not clear whether it will take a year+ for the speculated arrival of upgraded "Blackwell" desktop gaming solutions, but members of the Chiphell forum have been openly discussing alleged "in-progress" GeForce RTX 5080 SUPER 24 GB and RTX 5070 SUPER 18 GB card designs. Past weekend theorizations were highlighted by ITHome and VideoCardz—in particular, one Chiphell participant posited the two follow-ups will not be deployed with noticeably "faster" performance.

Larger pools of onboard GDDR7 VRAM could pave the way for improvements in AI productivity, although greater capacities could lead to beneficial conditions in gaming scenarios—e.g. extra provisions for large textures and complex assets. The speculated GeForce RTX 5080 SUPER card could borrow aspects from Team Green's GeForce RTX 5090 Mobile 24 GB model; namely the latter's usage of 3 GB GDDR7 memory modules. Press interpretations—of fresh Chiphell leaks—point to company engineers enabling existing PCB designs with 8×3 GB (24 GB) and 6×3 GB (18 GB) parts; thus preventing a major overhaul of board layouts. Global PC gaming hardware communities have often expressed a dislike of NVIDIA's repeated deployment of 8 GB and 12 GB capacity products. A theoretical GeForce RTX 5070 SUPER 18 GB option could receive a warmer welcome. Very early speculation proposes an introduction—of SUPER "Blackwell" models—just before or during CES 2026.



View at TechPowerUp Main Site | Source

 

[3DVCash]

Hmmmmm.... with a node shrink and boosted clocks this could get very interesting.

But ofc, none of this matters without pricing and availability.

 

[ZoneDymo]

Hmmmmm.... with a node shrink and boosted clocks this could get very interesting.

But ofc, none of this matters without pricing and availability.

in what universe do you live that after all we have seen for all these years you think there is even a chance of this being "interesting"?

 

[Peterson!]

Nvidia now that AI is a trend need to put in the RTX 6070 series 32GB memory, in RTX 6080 series 64GB memory and in RTX 5090 series,128GB memory for 6000 generation or else whats the point someone will buy?

 

[Marcus L]

Nvidia now that AI is a trend need to put in the RTX 6070 series 32GB memory, in RTX 6080 series 64GB memory and in RTX 5090 series,128GB memory for 6000 generation or else whats the point someone will buy?

if your main use case is for Ai/workstation purposes then pony up and buy a workstation class card, there's no way in hell leather jacket man will give you 32GB on a 70 class card for the next 10 years at least

 

[3DVCash]

in what universe do you live that after all we have seen for all these years you think there is even a chance of this being "interesting"?

Pure hopium, my friend! But we know Blackwell wasn't originally intended for N4... maaaaaaybe they correct that with the refresh.

If it's JUST a memory capacity bump, well... meh.

 

[freeagent]

Well.. I guess I will get one..

 

[Scattergrunt]

@AusWolf Thoughts?

To me this sounds pretty interesting. But its just a rumor.. for now. Hopefully stays a rumor.

 

[truerock]

I'm not doing much on my PC right now - email, web browsing.
I'm using 2GB of video memory on my Nvidia GeForce RTX 3050.

But, I do think 16GB of video memory is just about the new minimum for PC games.

 

[CrAsHnBuRnXp]

Why not just class this as a 5080Ti?

 

[AusWolf]

@AusWolf Thoughts?

To me this sounds pretty interesting. But its just a rumor.. for now. Hopefully stays a rumor.

Nothing surprising, imo. Nvidia has been pulling the "oh look, there's more" trick since Turing.

The only thing that mildly surprises me is that it's so soon after the generational launch, but considering that the cards aren't extremely popular, I guess they had to resort to extreme measures.

 

[GodisanAtheist]

They probably want to get the 5080 16Gb and 5070 12gb sorted out first before going to the next big thing.

 

[Event Horizon]

Someone here called it, and they've done this before. Launch with lower than expected VRAM and then refresh with the "right amount".

 

[docnorth]

IIRC both dies are already maxed out. Nvidia probably prefers the 3gb modules for a refresh instead of partially using a larger die.

@Event Horizon you are right, such comments began already from the 2000 series refresh.

 

[RejZoR]

Why? The existing "non super" cards are already stupid expensive and unobtainable. What can new SKU's possibly improve beyond being even more expensive and even more unobtainable? NVIDIA is stupid.

 

[OneSorcerer]

We could get significantly better value for money cards if supply is decent and UDNA delivers.

5060 Ti Super 128 bit 12GB for 449 street price (399 msrp), matching 3080 performance;
5070 Super 192 bit 18GB for 599 street price (549 msrp), matching 9070XT performance.
5070 Ti Super 192 bit 18GB for 799 street price (749 msrp), matching 5080 performance.
5080 Super 256 bit 24GB for 1000++ street price (1000+ msrp), matching 4090 performance.

 

[Sol_Badguy]

Very juicy subject. I like it, I like it a lot.

A full-die type Super is only possible with the GB205 (currently used on the 5070) and the GB202 (currently used on the 5090).
Regarding the GB202 I am willing to bet money that nVidia will not give such a unicorn to the consumer market.

But the GB205 is very likely. With perhaps a slight speed bump it will probably be a bit faster (in raster) than the 9070.
Landing at the same $550 MSRP as the 5070 it will replace means it could potentially be disruptive against the 9070 but not so much for the 9070 XT which will still be faster.
However regarding VRAM capacity, going from 12 to 18 would be most welcome, and going with 18 would definitely help the "5070 Super" be disruptive against the 9070 XT, although giving so much more at the same MSRP has a big question mark next to it.
nVidia sort of did it before with the 4070 Ti Super which represented a substantial upgrade over the 4070 Ti, compare this with the absolute joke of going from 4080 to 4080 Super, that was mostly a PR move for using the "MSRP reduction card".

Back to the 18GB dilemma, this move while very effective against 9070 series cards forces nVidia to make substantial VRAM upgrades further up the stack. An 18GB 5070 Super would make sense next to the 5060 Ti 16GB but would make the current 5070 Ti with 16GB look gimped, meaning the 5070 Ti needs at least 18GB or more to make sense in the lineup. For the 5070 Ti 18GB is possible with 3GB chips by enabling only 6 out of 8. Or perhaps 21GB by enabling 7 out of 8.
And so for the 5080 the possibility is with 21GB or 24GB. To give users 50% more VRAM at the same MSRP for the 5080 I imagine nVidia is foaming at the mouth.
And all of this thanks to "good guy" AMD.

The 5060 Ti is in a weird position, 3GB chips would make a 12GB version possible which has the advantage of having chips on only one side of the PCB and also 50% bump over the 8GB version but it would represent a buffer downgrade compared to the current 16GB version.
If this 12GB version gets the $380 MSRP then it would be an okay deal, but still coexisting with the 16GB version would be very weird.

There is another version, one in which nVidia is more stingy with the VRAM. So the 5070 Super gets only 15GB by having 5 out of 6 chips enabled. This means the 5070 Super won't be as disruptive against the 9070 series cards but this version is almost guaranteed to land at the same $550 MSRP as the 5070. Compared to the 5060 Ti 16GB (if that is still kept) it would look slightly gimped. But this means that the 5070 Ti could remain unchanged, or if nVidia wants to use 3GB chips across the stack then 18GB is the most likely. Then for the 5080 18GB is still possible but 21GB is more likely. There would be no reason that nVidia will go for 24GB.
In this version the lineup could look like this:
5060 Ti Semi-Super 12GB - $380 MSRP
5070 Super 15GB - $550 MSRP
5070 Ti Super 18GB - $750 MSRP
5080 Super 21GB - $1000 MSRP
3GB increments, only the 5060 Ti gets all chips enabled. 5060 Ti 16GB is discontinued and everything else higher up is replaced.
The lower-tier cards can still be countered by AMD's offer, maybe with a $50 price cut.

The rationale for the 50 series refresh is simply a testing ground for the 3GB GDDR7 chips, to see how they perform in dGPU consumer products in preparation for the 60 series. And of course to maintain the appearance of progress, of giving more for same or less money and all of that.
This is the sole reason in my opinion, there is no problem with how attractive the current cards are, the market is already adjusting all GPUs according to the meaningful variables: supply and demand, MSRP and actual performance. We see that the 9070 series were pushed upwards towards the 5070 Ti and not the other way around. We see the 5070 getting lower street price compared to the 9070.

 

[OneSorcerer]

Very juicy subject. I like it, I like it a lot.

A full-die type Super is only possible with the GB205 (currently used on the 5070) and the GB202 (currently used on the 5090).
Regarding the GB202 I am willing to bet money that nVidia will not give such a unicorn to the consumer market.

But the GB205 is very likely. With perhaps a slight speed bump it will probably be a bit faster (in raster) than the 9070.
Landing at the same $550 MSRP as the 5070 it will replace means it could potentially be disruptive against the 9070 but not so much for the 9070 XT which will still be faster.
However regarding VRAM capacity, going from 12 to 18 would be most welcome, and going with 18 would definitely help the "5070 Super" be disruptive against the 9070 XT, although giving so much more at the same MSRP has a big question mark next to it.
nVidia sort of did it before with the 4070 Ti Super which represented a substantial upgrade over the 4070 Ti, compare this with the absolute joke of going from 4080 to 4080 Super, that was mostly a PR move for using the "MSRP reduction card".

Back to the 18GB dilemma, this move while very effective against 9070 series cards forces nVidia to make substantial VRAM upgrades further up the stack. An 18GB 5070 Super would make sense next to the 5060 Ti 16GB but would make the current 5070 Ti with 16GB look gimped, meaning the 5070 Ti needs at least 18GB or more to make sense in the lineup. For the 5070 Ti 18GB is possible with 3GB chips by enabling only 6 out of 8. Or perhaps 21GB by enabling 7 out of 8.
And so for the 5080 the possibility is with 21GB or 24GB. To give users 50% more VRAM at the same MSRP for the 5080 I imagine nVidia is foaming at the mouth.
And all of this thanks to "good guy" AMD.

The 5060 Ti is in a weird position, 3GB chips would make a 12GB version possible which has the advantage of having chips on only side of the PCB and also 50% bump over the 8GB version but it would represent a buffer downgrade compared to the current 16GB version.
If this 12GB version gets the $380 MSRP then it would be an okay deal, but still coexisting with the 16GB version would be very weird.

There is another version, one in which nVidia is more stingy with the VRAM. So the 5070 Super gets only 15GB by having 5 out of 6 chips enabled. This means the 5070 Super won't be as disruptive against the 9070 series cards but this version is almost guaranteed to land at the same $550 MSRP as the 5070. Compared to the 5060 Ti 16GB (if that is still kept) it would look slightly gimped. But this means that the 5070 Ti could remain unchanged, or if nVidia wants to use 3GB chips across the stack then 18GB is the most likely. Then for the 5080 18GB is still possible but 21GB is more likely. There would be no reason that nVidia will go for 24GB.
In this version the lineup could look like this:
5060 Ti Semi-Super 12GB - $380 MSRP
5070 Super 15GB - $550 MSRP
5070 Ti Super 18GB - $750 MSRP
5080 Super 21GB - $1000 MSRP
3GB increments, only the 5060 Ti gets all chips enabled. 5060 Ti 16GB is discontinued and everything else higher up is replaced.
The lower-tier cards can still be countered by AMD's offer, maybe with a $50 price cut.

The rationale for the 50 series refresh is simply a testing ground for the 3GB GDDR7 chips, to see how they perform in dGPU consumer products in preparation for the 60 series. And of course to maintain the appearance of progress, of giving more for same or less money and all of that.
This is the sole reason in my opinion, there is no problem with how attractive the current cards are, the market is already adjusting all GPUs according to the meaningful variables: supply and demand, MSRP and actual performance. We see that the 9070 series were pushed upwards towards the 5070 Ti and not the other way around. We see the 5070 getting lower street price compared to the 9070.

I am trying to figure how a 15GB card would be possible...

15GB card, assuming 160 bit, would give 160/32 = 5, so we could have 10GB or 20GB without mismatching memory modules, and a 3GB module x 5 would leave one unpaired/disabled, reducing performance.
Assuming 192 bit, 192/32 = 6GB, so, 12GB or 18GB.
There will be no 15GB card?

 

[Sol_Badguy]

@OneSorcerer

I am not talking about mismatching 2GB with 3GB modules I am talking about enabling the modules.
Meaning there will be only 5 chips on the PCB instead of 6.

 

[OneSorcerer]

@Sol_Badguy But doesn't that reduce performance? Like it did with the gtx 970?

 

[Onasi]

I am not talking about mismatching 2GB with 3GB modules I am talking about enabling the modules.
Meaning there will be only 5 chips on the PCB instead of 6.

That would still end up being an asymmetrical configuration that cannot work on a 192-bit bus. You can’t just “disable” a memory chip and keep the same bus. I suppose theoretically it can be done in firmware to make only 15 gigs accessible while being physically and electrically present, just not technically addressable by applications, but that’s silly since you still end up paying for all 6 chips in manufacturing.

But doesn't that reduce performance? Like it did with the gtx 970?

The 970 was a different case. And there it only reduced performance if the last 0.5 gigs were used/called for. It was an all around poor decision on NV part and there is a reason why no GPU since went for asymmetric config in any way.

 

[Denver]

Just what was needed, an anemic 5070 with more VRAM than the 5080. Glorious...

But seriously, why are tech sites relying on discussions from a Chinese forum as their source now?

 

[OneSorcerer]

That would still end up being an asymmetrical configuration that cannot work on a 192-bit bus. You can’t just “disable” a memory chip and keep the same bus. I suppose theoretically it can be done in firmware to make only 15 gigs accessible while being physically and electrically present, just not technically addressable by applications, but that’s silly since you still end up paying for all 6 chips in manufacturing.


The 970 was a different case. And there it only reduced performance if the last 0.5 gigs were used/called for. It was an all around poor decision on NV part and there is a reason why no GPU since went for asymmetric config in any way.

ty!! I hope they release 5060 ti super 12GB matching a 3080 in raster for 399.

 

[Sol_Badguy]

@Onasi

I was under the impression that it does work. If not then my bad. I see others have thought about this.
But the good news then is that nVidia is forced to make more hefty VRAM bumps.
So if I was wrong then I'm glad I was.

 

[AusWolf]

Why? The existing "non super" cards are already stupid expensive and unobtainable. What can new SKU's possibly improve beyond being even more expensive and even more unobtainable? NVIDIA is stupid.

Media coverage.