Sunday, June 29th 2025

NVIDIA GeForce RTX 5070 Ti SUPER Planned with 24 GB of GDDR7 Memory

We have already covered that NVIDIA is planning a SUPER refresh of its GeForce RTX 50 series GPUs, with the most recent leak revealing the GeForce RTX 5070 SUPER, which is expected to feature a slight increase in CUDA core count and 18 GB of memory. However, NVIDIA is also overhauling its RTX 5070 Ti with the SUPER treatment, now enabling the RTX 5070 Ti SUPER to feature as much as 24 GB of GDDR7 memory with no change in CUDA core count. According to kopite7kimi, the RTX 5070 Ti SUPER will feature the same 8,960 CUDA core configuration as the GB203-350-A1 SKU, based on the PG147-SKU55 PCB. The only difference here is the VRAM surplus resulting from an additional 8 GB of GDDR7, which totals 24 GB capacity, compared to the 16 GB now on the regular RTX 5070 Ti SKU.

This also changes the TBP of the card, where the RTX 5070 Ti SUPER GPU is now rumored to feature a 350 W TBP, an upgrade from the non-SUPER's RTX 5070 Ti, which has a 300 W power rating. Perhaps the extra power is dedicated now not only to more VRAM (which is also a power consumer) but to higher clock speeds, too. It wouldn't be sufficient to call it a SUPER variant without some raw compute increase, so NVIDIA might be selecting better die bins for the SUPER SKU to enable higher default clock speeds at only a 50 W TBP increase. Nonetheless, we expect to hear more details in the coming weeks, especially since the wave of rumors has started to flood social media and leaks are becoming more frequent.
Source: kopite7kimi
Add your own comment

83 Comments on NVIDIA GeForce RTX 5070 Ti SUPER Planned with 24 GB of GDDR7 Memory

#51
Rover4444
Panther_SeraphinnVidia AMD and Intel would go, your doing professional work, use professional cards.
Luckily Intel's upcoming professional cards will actually be affordable. NVIDIA and AMD can pound sand, used 3090 here we go again...
Grimlingactually 24gig of vram should be standard by now for middle-class gpus, 16gig should be entry level.
i am struggling with 16gb right now using AI-models like the stable diffusion 3.5 big model (has like 28gig). so for me the 5080 should have at least 32gig, not 16, that is a joke.
Should be 33, it's triple the performance of the 1080 Ti after all :roll:

Would be funny if one of the companies went crazy and just stacked some CAMM2 or DDR modules on a board. Screw everything else, I want VRAM!
Posted on Reply
#52
Legacy-ZA
Anyways, the big thing we can take home from all of this is, try and hold out for the Super variants.
Posted on Reply
#53
Panther_Seraphin
Legacy-ZAAnyways, the big thing we can take home from all of this is, try and hold out for the Super variants.
Anyways, the big thing we can take home from all of this is, try and hold out for the Super variants AND PRAY THEY DONT UP THE COST


It was interesting to see HWUB do a 9070XT vs 5070ti re-review after a couple of months of fine wine aging from AMD.
Posted on Reply
#54
phints
What's the point of this nothing is going to use it. Wake me up when they increase the 8 and 12 GB cards. 18GB 5070 class will be decent.
Posted on Reply
#55
costabunny
Interesting if they actually go this way. Rumours are rumours, no matter the source.

Be nice to actually see a 70 series with 16 or more gigs of vram. I guess the prices will be the biggest factor when we eventually get confirmation from the big green monster.
Posted on Reply
#56
igormp
phintsWhat's the point of this nothing is going to use it. Wake me up when they increase the 8 and 12 GB cards. 18GB 5070 class will be decent.
Those would be pretty interesting for compute. I'd buy one myself, for the right price.
Posted on Reply
#57
CornyOC
This further proves that it's not worth to buy non Super versions unless you buy xx90 flagship. You will get the actual GPU when super comes out, the initial non-super versions are just de-tuned version with less memory. Typical mega corp sales tactics, make a new product, slightly make it worse (since you know people will buy), then release the real product a year later to sell even more to milk the generation.

Anyway, 8GB VRAM should've been ditched since 30xx RTX. 40xx and 50xx should've been 12GB at least. If the budget and mid range GPUs remain 8GB, game developers will have to choke their games so they will be able to run on those turds.
Posted on Reply
#58
Panther_Seraphin
CornyOCThis further proves that it's not worth to buy non Super versions unless you buy xx90 flagship. You will get the actual GPU when super comes out, the initial non-super versions are just de-tuned version with less memory. Typical mega corp sales tactics, make a new product, slightly make it worse (since you know people will buy), then release the real product a year later to sell even more to milk the generation.

Anyway, 8GB VRAM should've been ditched since 30xx RTX. 40xx and 50xx should've been 12GB at least. If the budget and mid range GPUs remain 8GB, game developers will have to choke their games so they will be able to run on those turds.
I can understand certain aspects due to the lack of viable 3Gb dies meaning either the memory controller has to be made a lot more complicated for the lower end products to hit the relevant memory capacity or it results in clamshell memory being used which has its own issues etc

but that is what you get for jumping 3 Memory techs in 3 generations.....DDR6 to 6X to 7
Posted on Reply
#59
Sol_Badguy
CornyOCThis further proves that it's not worth to buy non Super versions unless you buy xx90 flagship. You will get the actual GPU when super comes out, the initial non-super versions are just de-tuned version with less memory. Typical mega corp sales tactics, make a new product, slightly make it worse (since you know people will buy), then release the real product a year later to sell even more to milk the generation.

Anyway, 8GB VRAM should've been ditched since 30xx RTX. 40xx and 50xx should've been 12GB at least. If the budget and mid range GPUs remain 8GB, game developers will have to choke their games so they will be able to run on those turds.
I disagree regarding this current generation.

The 40 series was pretty special, the Supers came with a shader increase and a price reduction. The 4070 Ti Super also had a bus increase and subsequent VRAM buffer increase.
The 4070 Ti Super was the best value for that performance bracket, paying more for the 4080 Super wasn't really justified, this same situation is seen today with the 5070 Ti and the 5080.

But the 50 series Supers are rather dissapointing except the 5070 Super because that gets the full die and more VRAM (more than needed for gaming, but at least is above the threshold where in some situations 12GB might not suffice).
If it keeps the same MSRP then it will be a banner, probably replacing the 5060 Ti 16 GB as an entry level workstation card, well not that entry level in this case.

Whereas the 5070 Ti Super gets way more VRAM than it needs, so waiting for that instead of having bought the 5070 Ti 6 months ago is just stupid. The 5080 Super is definitely a better product than the 5080, so in that regard waiting for the 5080 Super instead of having bought the 5080 is justified.

The whole point of the refresh lineup is to make the early adopters feel shafted.
The 40 series managed this, the ones who bought the early lineup felt shafted when they saw the Supers.

The 50 series did not manage this.
Do you think the ones who bought the 4070 Super felt shafted because they haven't waited for the 5070? No.
Do you think the ones who bought the 4070 Ti Super felt shafted because they haven't waited for the 5070 ti? No.
Do you think the ones who bought the 4080 Super felt shafted because they haven't waited for the 5080? No.
Do you think the ones who bought the 4090 felt shafted because they haven't waited for the 5090? No. They sold their 4090, added some money and bought the 5090.

But no one who bought the 5070 Ti will feel shafted for not waiting for the 5070 Ti Super. Not for gaming anyway, maybe for AI.
The ones who bought the 5080 will feel shafted for not waiting for the 5080 Super (assuming the MSRP stays the same) but they already feel shafted for not buying the 5070 Ti and extra storage with the saved money.
The ones who bought the 5070 will feel the most shafted, especially if the 5070 Super keeps the same MSRP, or $600 at the most.

Also another dissapointing aspect is the fact the low tier cards don't get 3GB chips. The 5060 Ti is in a weird spot, relative to the 8GB model a 12GB one (using 3GB chips) is a proper upgrade, but compared to the 16GB one it wil be a downgrade.
The 5060 would also be weird to get 12GB, as it would partially cannibalize the hypothetical 5060 ti 12GB, which would be forced to keep the 16GB buffer in order to slot in between the 5060 12GB and the 5070 Super 18Gb.

But there's another issue, the 60 series. That generation will surely be something more substantial than the 40 -> 50 series uplift.
As such the ones who will buy the 50 Supers will feel the most shafted when the 60 series lands.
Therefore the wisest thing they can do is keep their money and grind it out until it's worth it.
Posted on Reply
#60
lexluthermiester
Sol_BadguyI disagree regarding this current generation.

The 40 series was pretty special, the Supers came with a shader increase and a price reduction. The 4070 Ti Super also had a bus increase and subsequent VRAM buffer increase.
The 4070 Ti Super was the best value for that performance bracket, paying more for the 4080 Super wasn't really justified, this same situation is seen today with the 5070 Ti and the 5080.

But the 50 series Supers are rather dissapointing except the 5070 Super because that gets the full die and more VRAM (more than needed for gaming, but at least is above the threshold where in some situations 12GB might not suffice).
If it keeps the same MSRP then it will be a banner, probably replacing the 5060 Ti 16 GB as an entry level workstation card, well not that entry level in this case.

Whereas the 5070 Ti Super gets way more VRAM than it needs, so waiting for that instead of having bought the 5070 Ti 6 months ago is just stupid. The 5080 Super is definitely a better product than the 5080, so in that regard waiting for the 5080 Super instead of having bought the 5080 is justified.

The whole point of the refresh lineup is to make the early adopters feel shafted.
The 40 series managed this, the ones who bought the early lineup felt shafted when they saw the Supers.

The 50 series did not manage this.
Do you think the ones who bought the 4070 Super felt shafted because they haven't waited for the 5070? No.
Do you think the ones who bought the 4070 Ti Super felt shafted because they haven't waited for the 5070 ti? No.
Do you think the ones who bought the 4080 Super felt shafted because they haven't waited for the 5080? No.
Do you think the ones who bought the 4090 felt shafted because they haven't waited for the 5090? No. They sold their 4090, added some money and bought the 5090.

But no one who bought the 5070 Ti will feel shafted for not waiting for the 5070 Ti Super. Not for gaming anyway, maybe for AI.
The ones who bought the 5080 will feel shafted for not waiting for the 5080 Super (assuming the MSRP stays the same) but they already feel shafted for not buying the 5070 Ti and extra storage with the saved money.
The ones who bought the 5070 will feel the most shafted, especially if the 5070 Super keeps the same MSRP, or $600 at the most.

Also another dissapointing aspect is the fact the low tier cards don't get 3GB chips. The 5060 Ti is in a weird spot, relative to the 8GB model a 12GB one (using 3GB chips) is a proper upgrade, but compared to the 16GB one it wil be a downgrade.
The 5060 would also be weird to get 12GB, as it would partially cannibalize the hypothetical 5060 ti 12GB, which would be forced to keep the 16GB buffer in order to slot in between the 5060 12GB and the 5070 Super 18Gb.

But there's another issue, the 60 series. That generation will surely be something more substantial than the 40 -> 50 series uplift.
As such the ones who will buy the 50 Supers will feel the most shafted when the 60 series lands.
Therefore the wisest thing they can do is keep their money and grind it out until it's worth it.
So to sum this up, the RTX5000 series did the same things for RTX4000 that the GTX500 did for GTX400. Some improvements, but just an inta-generational uptick. NVidia called it Blackwell instead of Ada+ but it's still just incremental.
Posted on Reply
#61
Sol_Badguy
lexluthermiesterSo to sum this up, the RTX5000 series did the same things for RTX4000 that the GTX500 did for GTX400. Some improvements, but just an inta-generational uptick. NVidia called it Blackwell instead of Ada+ but it's still just incremental.
Pretty much, but with some differences.
From what I remember people were clearly more satisfied with the GTX 500 cards than the GTX 400 ones. The better temps were no doubt the main factor, but the cards themselves were very well balanced, overclockable, and of course the GTX 580 reigned supreme with no AMD opponent.

Also because the arch was essentially Fermi there were no functional differences (if I'm not mistaken) between the generations, so all of the graphics features (DirectX, OpenGL, CUDA, Shader Model etc.) were identical, whereas Blackwell has some architectural differences, and is better optimized for AI and whatever (this can be seen as a positive for Blackwell, although it's not something significant).

On top of that there was less than year between those generations, but the 5090 came after two years and three months after the 4090.
And in this example we get the largest performance gap.

So I feel that despite the rehash similarity between Fermi->Fermi 2.0 and Ada->Blackwell, the GTX 500 was a necessary and very welcome improvement that came almost immediately, but the RTX 50 series is just there to buy time until 60 series. A stop gap, filler, nothing more.

Well it is more than that actually, it's also a missed opportunity for AMD. I mean they made REAL progress with RDNA4, yet there's no bigger chip, and now they know as well as we that it was a mistake to hold their punches, the bigger chip would've have sold (the full die and the gimped runner-up), 100% it would have sold, people are definitely not impressed by the 50 series, and now with the 5080 Super they're still not quite getting that 4090 performance for $1K (fingers crossed it's not 1200) after what 3 years?
Posted on Reply
#62
Panther_Seraphin
I dont understand what nVidia are doing with all the partially defective 5090 dies

Those things are MASSIVE so you arent selling it to me they have 100% viable or completely dead dies. There has to be a lot that could have been harvested into something between the existing 5080 and 5090 offerings
Posted on Reply
#63
Sol_Badguy
@Panther_Seraphin

www.techpowerup.com/gpu-specs/nvidia-gb202.g1072
The most cut down is the RTX Pro 5000 Blackwell, which has less SM units than the 4090, go figure.
But even the most hardcore RTX Pro 6000 Blackwell Server doesn't have the full die.
Probably because they get so few that they won't have enough to justify marketing a new product, the demand would obviously be huge much bigger than how many full dies they could get so most likely they gimp the full ones into the Pro 6000, so that way the supply sort of keeps up with the demand, I don't know.
But surely from the 5090 to the Pro 5000 is too big of a gap, a lot of dies have to end up in that zone, so probably they just gimp many of them and still sell them for profit in the form of the Pro 5000, which costs way more than the 5090.

Making a GPU that slots in between the 5080 and the 5090 is less profitable, considering it wouldn't have less processing units than the Pro 5000 but it could also have more, like the 4090, and to have an MSRP of $1600.

Actually about what I said above, they might also put some dies on the side, and next year when the Supers launch, they could have a new Pro card as well to slot in between the 5000 and the 6000. As far as yields go, accumulating them by then they would have enough to meet demand.
But I don't think this will happen though, seems a bit too good for the enterprise market, and good business means no one should get exactly what they want, gotta keep 'em hungry for more.
Posted on Reply
#64
Panther_Seraphin
Sol_Badguy@Panther_Seraphin

www.techpowerup.com/gpu-specs/nvidia-gb202.g1072
The most cut down is the RTX Pro 5000 Blackwell, which has less SM units than the 4090, go figure.
But even the most hardcore RTX Pro 6000 Blackwell Server doesn't have the full die.
Probably because they get so few that they won't have enough to justify marketing a new product, the demand would obviously be huge much bigger than how many full dies they could get so most likely they gimp the full ones into the Pro 6000, so that way the supply sort of keeps up with the demand, I don't know.
But surely from the 5090 to the Pro 5000 is too big of a gap, a lot of dies have to end up in that zone, so probably they just gimp many of them and still sell them for profit in the form of the Pro 5000, which costs way more than the 5090.

Making a GPU that slots in between the 5080 and the 5090 is less profitable, considering it wouldn't have less processing units than the Pro 5000 but it could also have more, like the 4090, and to have an MSRP of $1600.

Actually about what I said above, they might also put some dies on the side, and next year when the Supers launch, they could have a new Pro card as well to slot in between the 5000 and the 6000. As far as yields go, accumulating them by then they would have enough to meet demand.
But I don't think this will happen though, seems a bit too good for the enterprise market, and good business means no one should get exactly what they want, gotta keep 'em hungry for more.
Oh I know,

I just love the fact that of every generation bar this one the highest end die was reused in lower end consumer parts depending on yields and defects in the main die. It just reeks of profiteering by basically saying pay top dollar for a 5090 or pay maybe 50-60% for exactly 50% of a 5090.

There is so much space for Ti's, Supers AND super Ti's in the stack with the GB202 it just shows that nVidia sees the gaming division as the peasants now and will only get the scraps off the table as and when they fall off it and not a minute sooner.
Posted on Reply
#65
Rover4444
Sol_BadguyI disagree regarding this current generation.

The 40 series was pretty special, the Supers came with a shader increase and a price reduction. The 4070 Ti Super also had a bus increase and subsequent VRAM buffer increase.
The 4070 Ti Super was the best value for that performance bracket, paying more for the 4080 Super wasn't really justified, this same situation is seen today with the 5070 Ti and the 5080.

But the 50 series Supers are rather dissapointing except the 5070 Super because that gets the full die and more VRAM (more than needed for gaming, but at least is above the threshold where in some situations 12GB might not suffice).
If it keeps the same MSRP then it will be a banner, probably replacing the 5060 Ti 16 GB as an entry level workstation card, well not that entry level in this case.

Whereas the 5070 Ti Super gets way more VRAM than it needs, so waiting for that instead of having bought the 5070 Ti 6 months ago is just stupid. The 5080 Super is definitely a better product than the 5080, so in that regard waiting for the 5080 Super instead of having bought the 5080 is justified.

The whole point of the refresh lineup is to make the early adopters feel shafted.
The 40 series managed this, the ones who bought the early lineup felt shafted when they saw the Supers.

The 50 series did not manage this.
Do you think the ones who bought the 4070 Super felt shafted because they haven't waited for the 5070? No.
Do you think the ones who bought the 4070 Ti Super felt shafted because they haven't waited for the 5070 ti? No.
Do you think the ones who bought the 4080 Super felt shafted because they haven't waited for the 5080? No.
Do you think the ones who bought the 4090 felt shafted because they haven't waited for the 5090? No. They sold their 4090, added some money and bought the 5090.

But no one who bought the 5070 Ti will feel shafted for not waiting for the 5070 Ti Super. Not for gaming anyway, maybe for AI.
The ones who bought the 5080 will feel shafted for not waiting for the 5080 Super (assuming the MSRP stays the same) but they already feel shafted for not buying the 5070 Ti and extra storage with the saved money.
The ones who bought the 5070 will feel the most shafted, especially if the 5070 Super keeps the same MSRP, or $600 at the most.

Also another dissapointing aspect is the fact the low tier cards don't get 3GB chips. The 5060 Ti is in a weird spot, relative to the 8GB model a 12GB one (using 3GB chips) is a proper upgrade, but compared to the 16GB one it wil be a downgrade.
The 5060 would also be weird to get 12GB, as it would partially cannibalize the hypothetical 5060 ti 12GB, which would be forced to keep the 16GB buffer in order to slot in between the 5060 12GB and the 5070 Super 18Gb.

But there's another issue, the 60 series. That generation will surely be something more substantial than the 40 -> 50 series uplift.
As such the ones who will buy the 50 Supers will feel the most shafted when the 60 series lands.
Therefore the wisest thing they can do is keep their money and grind it out until it's worth it.
The 5070 Ti Super will have as much VRAM as the 3090 Ti and much better performance. There's finally a true 3090 replacement (no NVLink though) in the midrange, I don't think it's "way more VRAM than it needs". Six months ago 5070 Tis weren't at MSRP or weren't in stock either.

8GB 5060 Ti model should go out of production and the 5060 should get the new modules.
Posted on Reply
#66
Sol_Badguy
Panther_SeraphinThere is so much space for Ti's, Supers AND super Ti's in the stack with the GB202
This ties in with what I said above about AMD not making a bigger chip, a successor to 7900 XTX would've equaled or came really close to the 4090 level of performance and its gimped brother would've equaled or came really close to the 5080.

Yeah, yeah it's been talked about how the power draw would've been higher than opposing nVidia cards, how the feature set would've been less, how this and that, but at $1000 and $800 these things would've made a difference, and would've forced nVidia to start carving GB202 dies into stronger products than the 5080.

That's what you get when nVidia is unchallenged, they shift into 4th gear because they win anyway.

Daniel Owen made this video some time ago, good to remember when we get the 60 series, see where that lands:
Rover4444The 5070 Ti Super will have as much VRAM as the 3090 Ti and much better performance. There's finally a true 3090 replacement (no NVLink though) in the midrange, I don't think it's "way more VRAM than it needs". Six months ago 5070 Tis weren't at MSRP or weren't in stock either.

8GB 5060 Ti model should go out of production and the 5060 should get the new modules.
24GB for gaming is absurdly high for the amount of muscle that GPU has (the 4090 doesn't have enough muscle for 24GB in gaming).
For AI it's very welcome, although that kind of task would benefit from even more like 48GB, why not? The sky is the limit in that case, why do we have unrealistic demands from gaming cards? If they're optimized primarily for gaming then that's the lense through which we should judge them.

I understand the angle of comparing it with the 3090 Ti, you're now getting a midrange product that doesn't have any handicap compared to a flagship from two generations ago.
But again, for gaming where exactly is the current 5070 Ti limited compared to the 3090 Ti?

I'm the last person that would defend nVidia but they had to start somewhere with the 50 series, so they decided the VRAM buffers stay the same but the memory will be GDDR7. In time the VRAM buffers will increase, sometimes more than it fits the product so to speak.
You do agree that the 5070 Ti Super only gets the uplift because there was no other choice, right? If the 5080 Super gets 24GB and the 5070 Super gets 18GB then in between these there's only one possibility.

In my view the 5080 Super will be the better deal (in reverse of the current ones), in this case the price difference is much easier to swallow, the VRAM is higher speed, and the GPU has more muscle, not enough for the buffer, but still more, and that means the expiry date is likely a pinch further away.
But I still recommend holding the line until 60 series.
Posted on Reply
#67
Rover4444
Sol_BadguyFor AI it's very welcome, although that kind of task would benefit from even more like 48GB, why not? The sky is the limit in that case, why do we have unrealistic demands from gaming cards? If they're optimized primarily for gaming then that's the lense through which we should judge them.
It's not an unrealistic demand, AI development and usage of most models is largely driven by consumer cards.
Sol_BadguyBut again, for gaming where exactly is the current 5070 Ti limited compared to the 3090 Ti?
In VRAM. VR games and flight sims need it.
Sol_BadguyYou do agree that the 5070 Ti Super only gets the uplift because there was no other choice, right? If the 5080 Super gets 24GB and the 5070 Super gets 18GB then in between these there's only one possibility.
I honestly don't know why NVIDIA is giving the 5070 Ti Super an uplift, probably so they're more competitive with the 9070 XT if the MSRP remains the same or so it doesn't have less VRAM than the 5070 Super or so the entire lineup uses the new VRAM modules or something. The 50W power increase wouldn't be doing much for it otherwise in my opinion, and faster 2GB modules doesn't make sense when they can just use the slower 3GB modules that didn't make the cut for the 5080 Super. I am glad that they're doing this uplift though, I'm not going to knock them for it because the reduction in costs for skimping a bit on the VRAM are NOT going to be passed down to me.
Sol_BadguyIn my view the 5080 Super will be the better deal (in reverse of the current ones), in this case the price difference is much easier to swallow, the VRAM is higher speed, and the GPU has more muscle, not enough for the buffer, but still more, and that means the expiry date is likely a pinch further away.
But I still recommend holding the line until 60 series.
Meh, I think of the 5080 Super as more of a 4090 alternative. I'm getting the 5070 Ti Super because it'd be worth it for me.
Posted on Reply
#68
lexluthermiester
Sol_Badguy(the 4090 doesn't have enough muscle for 24GB in gaming)
That's not true at all. How much RAM a game needs to store data is not directly related to GPU performance. Remember, not everything stored in VRAM is actually being displayed on screen. VRAM is used for more than just the screen buffer.
Posted on Reply
#69
Sol_Badguy
@lexluthermiester

Well okay then where do you draw the line?
In order to say "oh the 4090 only has muscle for 20GB of VRAM, or 24GB or 28GB" (generally speaking, of course it varies per game).
The more you give it the more it stuffs with whatever background shit, but it won't translate into extra performance, that's why my approach is to identify the threshold where a lower amount of VRAM starts to affect performance then back up by several GB (let's say 4) and call it a day.

Looking in the Game Testing section here (I looked at all the games tested in 2025 for now) the results are varied, obviously not all the games use UE5 and not all that do have similar fps numbers. Also the games that use other engines aren't necessarily more forgiving of GPUs than UE5 titles, some are but not all.

The idea is that if the 4090 has less than 60 fps on 2160p maxed out in whatever game but the VRAM usage is at best below 16GB, then it's clear, it doesn't have enough muscle. Furthermore, if it scores 80-90 fps in another game but the VRAM usage is like at the most 12GB, what's the point in having tons of VRAM for gaming? Sure you could say that if you increase the resolution further until you drop close to 60 fps then the VRAM usage would increase, getting perhaps in the 20GB region, and then that would prove that the 4090 has enough muscle for the 24GB buffer, sort of (in those games at least), I mean you could play the game at 60 fps (although the type of game influences what fps number offers a good experience), but I doubt the experience is comparable to 2160p and 90 fps.
Again I'm not trying to cherry pick, in my opinion a more fluid native framerate should always trump chasing the maximum resolution, the highest detail settings and the most absurd texture packs. That's why a lot of people prefer using upscaled 2160p instead of native, less strain, less power draw, almost indiscernible visual difference.

A good opportunity to carry out extensive testing would have been a 5060 Ti 12 GB (with 3GB chips), because then we would have the same GPU with three different buffers. And it would allow far better pinpointing of what exactly is the optimal buffer for that GPU horsepower. And even though it would be a stretch to extrapolate those results to stronger GPUs at least it would be a valuable dataset, and with time other examples would help fill in more blanks regarding this subject.

Anyway that's my take on it.

@Rover4444

Okay but by the same logic that would also make the 7900 XTX superior (at least in some regards) than every other card with lower VRAM in that performance bracket, right?
Posted on Reply
#70
Rover4444
Sol_BadguyOkay but by the same logic that would also make the 7900 XTX superior (at least in some regards) than every other card with lower VRAM in that performance bracket, right?
Yes, of course.

Also if you're going to argue against the 4090 having 24GB of VRAM because it doesn't have "enough muscle" then that'd be pretty insane. It has over three times the performance as the 1080 Ti and 3060 as it is.
Posted on Reply
#71
lexluthermiester
Sol_BadguyWell okay then where do you draw the line?
That's impossible to do. This is because of the varied nature of a multitude of games being developed in different ways by different people. There is no way to definitively gauge such a metric.
Posted on Reply
#72
Panther_Seraphin
Sol_Badguy
Sol_Badguy*SNIP*
The main problem though is that VRAM usage has been on an upward trend in recent years

So even though your playing at low resolutions the game tends load up massive textures as that is all that is available and suddenly you VRAM usage has shot up for no reason. This is especially true in modded games, skyrim comes to mind.

When you look at the overall "cost saving" by not using the largest modules available its purely for the benefit of manufacturers and limits the longevity of cards especially for the prices they are charging currently. I understand currently the cards were given the memory they were at the time as the 3Gb modules were not in mass production at the time, its now possible to source 3Gb modules as an end user from around 3 months ago so you can guarantee nVidia/AMD would have been able to source them 6 months ago at least.

Yet Crickets.....
Posted on Reply
#73
Sol_Badguy
Rover4444Also if you're going to argue against the 4090 having 24GB of VRAM because it doesn't have "enough muscle" then that'd be pretty insane. It has over three times the performance as the 1080 Ti and 3060 as it is.
Well if I compare the 3090, 3090 Ti, 5070 Ti Super, 5080 Super, 4090, 6070 Ti and the 6080 all which have/will have 24GB and which don't have the same horsepower (I've listed them from weakest to strongest) it becomes apparent that at least one card in this list is underpowered relative to the buffer compared to the card at the other end.

I'm not advocating for less VRAM, truth be told the flagships always get a bit more than needed as a statement, the halo product is always used to make a statement, and the 5090 definitely took things to the next level.

Again I'm limiting my view to gaming, if we bring other tasks (AI) into the equation I feel we're skewing the perspective and these cards would end up sort of a middle ground between pure gaming and pure AI as far as VRAM buffers go -> say for gaming 20 GB would be just right for a particular card, but the AI crowd would want it to have 40 and the final product ends up having 30 to satisfy most people.
lexluthermiesterThat's impossible to do. This is because of the varied nature of a multitude of games being developed in different ways by different people. There is no way to definitively gauge such a metric.
Absolutely that's why I was talking about the situation encountered most often, there are always exceptions but you can't calibrate your product to satisfy those unless it's a flagship.
Panther_SeraphinWhen you look at the overall "cost saving" by not using the largest modules available its purely for the benefit of manufacturers and limits the longevity of cards especially for the prices they are charging currently. I understand currently the cards were given the memory they were at the time as the 3Gb modules were not in mass production at the time, its now possible to source 3Gb modules as an end user from around 3 months ago so you can guarantee nVidia/AMD would have been able to source them 6 months ago at least.
Of course but why would the GPU makers bend over backwards to give customers the ideal products? Again I'm not defending nVidia but AMD gave us the 9070 XT which in truth is a successor to 7800 XT, and it's definitely an upgrade BUUUT at an increased MSRP and with the same VRAM buffer and with GDDR of the same kind!
So if AMD held its punches why would nVidia be all generous? They're not Santa Clause.

Everybody got played this turn, come to think of it given Blackwell is a sort of remix of Ada I'm wondering if the Blackwell arch wasn't actually made by nVidia back in 2022 and they gimped it on purpose, called it Ada and released it in that state because it was enough to win against AMD, it was enough to have a significant uplift over 30 series etc. And they left some juice in the tank to squeeze later with 50 series. Tick-tock, incremental uplifts.
Posted on Reply
#74
lexluthermiester
Sol_BadguyAbsolutely that's why I was talking about the situation encountered most often
Then you are arguing against logic. Most games are not that heavy but many are now using as much VRAM as possible to improve/increase render distance and quality of rendered distant objects as well has keeping environmental assets loaded in VRAM for instant use. This dynamic is not compute heavy but is VRAM heavy. Thus a 3090 is still enough for most games that push VRAM usage to the limits. A 4090 will have no issues with the vast majority of titles.
Posted on Reply
#75
Panther_Seraphin
Sol_BadguyOf course but why would the GPU makers bend over backwards to give customers the ideal products? Again I'm not defending nVidia but AMD gave us the 9070 XT which in truth is a successor to 7800 XT, and it's definitely an upgrade BUUUT at an increased MSRP and with the same VRAM buffer and with GDDR of the same kind!
So if AMD held its punches why would nVidia be all generous? They're not Santa Clause.

Everybody got played this turn, come to think of it given Blackwell is a sort of remix of Ada I'm wondering if the Blackwell arch wasn't actually made by nVidia back in 2022 and they gimped it on purpose, called it Ada and released it in that state because it was enough to win against AMD, it was enough to have a significant uplift over 30 series etc. And they left some juice in the tank to squeeze later with 50 series. Tick-tock, incremental uplifts.
Unfortunately we were screwed out of decent competition in the high end when Big Navi was deemed unviable.

My guess is they tried to do what they are doing for the next Gen CPU but on the GPU earlier and it was just a step to far too soon and it would most likely have to be the tricker stacking of the Memory Interconnects below the GCD die
Posted on Reply
Add your own comment
Nov 17th, 2026 21:10 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts