Nvidia RTX 5000 Super cards could increase VRAM by 50%, with 24GB RTX 5080 Super and 18GB RTX 5070 Super

[midian182]

Rumor mill: There's little doubt that Nvidia will at some point release Super variants of its RTX 5000 series, the question is whether they will address some of the issues that have made consumer Blackwell one of the worst-received graphics architectures ever. The stingy amount of VRAM has been one of the many complaints directed against the cards, but rumors say that this will improve in the Super versions.

According to claims on China's Chiphell forums, Nvidia is preparing an RTX 5080 Super with 24GB of VRAM, up from the vanilla version's 16GB. That would match the memory size of the RTX 4090, 3090, and 3090 Ti.

Nvidia is also said to be readying an RTX 5070 with 18GB, a 6GB increase compared to the standard card's 12GB. This would be a new memory configuration for Nvidia as none of its previous cards have 18GB of VRAM.

NVIDIA is preparing RTX 50 SUPER products?
- RTX 5080 SUPER 24GB
- RTX 5070 SUPER 18GBhttps://t.co/JIQtp65hGrhttps://t.co/VhW5DdIeCs

– 포시포시 (@harukaze5719) April 28, 2025

While the cards will stick with the same type of GDDR7 VRAM as the standard versions, it's believed that they will use 3GB memory modules to enable the non-standard configurations.

While all rumors of this kind, especially those from the Chiphell forums, should be taken with a healthy dose of salt, there are some elements that lend credence to the claims, including the fact that the RTX 5090 laptop GPU uses 3GB modules and has a 24GB capacity.

Moreover, Nvidia reportedly planned on releasing a 24GB RTX 5080 but the cost and limited availability of 3GB modules resulted in it using a 16GB configuration – we even saw promo material from MSI that showed a 24GB RTX 5080, though it could have been an error on the company's part or even a fake. There were also rumors earlier this year of a 24GB RTX 5080 variant being in the works.

Courtesy of VideoCardz

A 50% increase in the cards' memory capacity would be welcome, especially in the case of the RTX 5070, which, as we mentioned in our review, is using a 12GB memory capacity that should really be reserved for entry-level products.

However, according to one Chiphell user, the performance of the new Super models doesn't feel that much faster than the current RTX 5000 equivalents. That would line up with the previous generation Super models: the RTX 4080 Super is just 2% faster than the RTX 4080 at 1080p, 3% faster at 1440p, and 4% faster at 4K.

Only two desktop Super graphics cards have had more memory than their standard counterparts: the RTX 2060 (6GB) / 2060 Super (8GB), and the RTX 4070 Ti (12GB) / RTX 4070 Ti Super (16GB). Again, we found the RTX 4070 Ti Super to be only around 3% to 5% faster than the RTX 4070 Ti.

It was about 15 months after the RTX 4000 series launched when the Super variants were announced. If Nvidia sticks with the same release schedule, the RTX 5000 Super cards won't arrive until the middle of next year.

Permalink to story:

Nvidia RTX 5000 Super cards could increase VRAM by 50%, with 24GB RTX 5080 Super and 18GB RTX 5070 Super

 

[Loadedaxe]

These memory configurations are what they should have started with at the current price scheme with lower pricing for the current cards on the market.

 

[opckieran]

Even Apple capitulated and bumped up their minimum RAM spec to 16GB on their traditional compute devices. Apple. Come on, nVidia…

 

[yRaz]

They need a price drop along side this. The super series brought back some good will with the 40 series, hopefully this will do the same. I might hate on then a lot, but I have no desire for their them to be making over priced products or taking the market in the direction it's going.

 

[Neatfeatguy]

That 5080 Super is going to give the 5080 a run for it's money! Just like the 4080 Super did to the 4080. That roughly 2% performance gain was off the chart!

The only difference this time around is the 5080 Super is going to be more expensive due to adding more RAM. The fake MSRP will be $1499, in reality they'll be at the $2k+ range.

 

[ScottSoapbox]

The 5080 Super can't be much faster as it is already using the full die so all that is left is increasing the core and memory clocks.

Extra VRAM will help it's longevity but you will also be waiting another year to start using it so...

...that basically leaves the hope for a price cut to make waiting really worth it. Maybe that happens? But maybe that is offset by tariffs? At this point, I'll probably still snag a 5080 if they get to MSRP rather than hope for the best in the future.

 

[GodisanAtheist]

5080 Ti/Super can always use a cut down 5090 die...

 

[Squid Surprise]

The supers are rarely a good buy - they’re simply Nvidia trying to stretch out the longevity of an existing architecture.

If you’re willing to wait 18 months, wait another 18 and get the 6000 series…

 

[Puiu]

Planned obsolescence in plain view.

 

[umeng2002]

Early adopters ripped off again

 

[brucek]

If you’re willing to wait 18 months, wait another 18 and get the 6000 series…

That was my strategy this past generation. Not sure the 5000 series was worth the wait.

 

[Squid Surprise]

That was my strategy this past generation. Not sure the 5000 series was worth the wait.

Well, with Nvidia, you have to do a bit of research... if their node hasn't changed, then you can't expect much in performance improvements... The 6000 series should be on a new node, so hopefully this means actual performance improvements... of course, you never really know until they're released.

 

[passwordistaco]

Are any of the Blackwell cards bandwidth-starved? It seems they could have kept costs down by using GDDR6/X on everything below the 5090.

 

[Lionvibez]

These memory configurations are what they should have started with at the current price scheme with lower pricing for the current cards on the market.

They already addressed this in the article.

"limited availability of 3GB modules"

that is out of NV's control as they don't make the memory modules!

 

[brucek]

"limited availability of 3GB modules"

that is out of NV's control as they don't make the memory modules!

Yes but maybe also a little no. Just how hard did Nvidia try to obtain this supply? I mean I agree these parts are not in common use elsewhere and therefore hardly falling off over-brimming warehouse bins, but it's always a chicken and egg situation where the industry produces what their customers are demanding. Did Nvidia's purchasing agents try their best and mention their gamer customers would probably be happy to offer 150% or 200% over asking for these parts? Or did they take the first answer they got which may have been what they wanted for the moment anyway?

This also all leaves out what alternate memory designs may have been possible, which is above my pay grade, but given the very high real world market price of the upper end cards I feel like there may have been room to have the memory people were looking for, if not for a desire to artificially limit it until a competitor forces their hand.

 

[daffy duck]

Lipstick on a broken pig unless the whole card is redesigned from the ground up.

 

[Hooda Thunkett]

I thought the companies branding the cards were making them, not Nvidia. Am I wrong about that? If that's the case, and current generation 5000 series chips can handle 3gb memory modules, it seems the card manufacturers could use them, right? Would Nvidia write the drivers to not work with those modules? Is there an insane hardware hacker out there willing to try to add 3gb modules to these cards to see if they work?

 

[Puiu]

Are any of the Blackwell cards bandwidth-starved? It seems they could have kept costs down by using GDDR6/X on everything below the 5090.

Without the faster memory you can imagine the little performance gains they have over the 4000 series would dwindle even more, unless they change the bus width to compensate.

 

[LimyG]

That was my strategy this past generation. Not sure the 5000 series was worth the wait.

Same. My last card purchase was the RTX 3080ti (Asus Strix gaming OC). It is still very good today. It also allows 450watt, and really uses it. (Default is 400watt max)

Did the same with GTX1xxx series. That was REALLY worth it. Ended up with the excellent GTX 1080ti.

Yes, means waiting. But hardware issues, if any, and more importantly drivers are mature by the time the supers and ti's come out.

As for the RTX 5080 total rip off, even some hardware issues (missing Rops) and trash drivers.

Depending on the pricing, I will buy the super or ti if it reviews well. With the extra VRAM I think the card would be good. The current RTX5080 wasn't even a consideration.

Hoping for sensible pricing.

 

[Lionvibez]

Yes but maybe also a little no. Just how hard did Nvidia try to obtain this supply? I mean I agree these parts are not in common use elsewhere and therefore hardly falling off over-brimming warehouse bins, but it's always a chicken and egg situation where the industry produces what their customers are demanding. Did Nvidia's purchasing agents try their best and mention their gamer customers would probably be happy to offer 150% or 200% over asking for these parts? Or did they take the first answer they got which may have been what they wanted for the moment anyway?

This also all leaves out what alternate memory designs may have been possible, which is above my pay grade, but given the very high real world market price of the upper end cards I feel like there may have been room to have the memory people were looking for, if not for a desire to artificially limit it until a competitor forces their hand.

lol there gamer customers?

NV doesn't give a damn about them they will continue to give you the scraps from their server and workstation products and you are going to like it.

Gamers think there way more important than they actually are.

 

[brucek]

NV doesn't give a damn about them they will continue to give you the scraps from their server and workstation products and you are going to like it.

So you're agreeing with me that this was not about the impossibility of obtaining 3GB modules, which was the point I was making.

 

[Puiu]

So you're agreeing with me that this was not about the impossibility of obtaining 3GB modules, which was the point I was making.

Nvidia is limited by the number of GPU dies they can produce, not VRAM modules.

It's all just planned obsolescence so that they can increase sales of future cards. It's also a business strategy to protect their high end AI market sales where VRAM is very important.

 

[Tran Bronstein]

The added VRAM wouldn't boost the actual GPU's performance per se -- it would make the graphics cards actually capable of running modern games that need the extra memory to properly fully load all the needed assets and textures. I would feel extremely bad for those who already bought RTX 5080s and 5070s though as such a release would man they got screwed by NVIDIA as payment for being loyal customers and early adopters.