NVIDIA GeForce RTX 5080 SUPER Could Feature 24 GB Memory, Increased Power Limits

[TheGeekn°72]

But there are no 192 bits this gen from AMD and Nvidia. That’s what’s so weird. Nvidia especially is known for 192, 320, 384 and sometimes even 448-bit buses in the past. Not one of those this gen even with the supers as they are using 3GB dies to offer mid memory sizes on 128- and 256-bit only.

It makes sense. 128-, 192, 256 are cheap and easy to make as I understand it, so why spend 512-bit memory interfaces on gaming plebs when majority is still gaming at 1080p, when you can sell it to datacenters? Well you can sell 512-bit junky dies in form of 4090/5090 to enthusiast gamers with deep pockets while selling good ones rebranded to A6000 labels for +5,000 bucks to enterprise and governments.

I wonder, is UDNA planned to be chiplet based ? because if they do something like Ryzen with a common IO die across SKUs, we could see larger buses if the economics of doing so are interesting enough for them to go with that... The "golden standard" is essentially the 192-256 bit bracket but honestly, just like with PCIe lanes on CPUs, I wouldn't mind if the amount we got was bumped up to 384-512...

 

[RedelZaVedno]

View attachment 405959

It'd be more accurate to say that Nvidia's gaming segment died on us. [Before anyone else mentions it, I'm not talking about sales.]
They released the waste of sand that is the 5050 like it's not subpar on every levels, the 3060 beats the 5060 as soon at the VRAM ceiling is hit, the 5080 is a 4080Ti and Blackwell's IPC uplift is essentially margin of error, if anything came out of RTX50, it's a massive AI uplift that nobody asked for and fast VRAM that no one who buys those cards will actually fully profit from, those who benefit from it are studios who can't be bothered to optimize their games and use DLSS as a crutch for their garbage and people buying RTX50 are the actual suckers paying for it.

Case in point : Hell Is Us requires permanently enabled upscaling and a 4090 to get 4K@30, what the actual fuck ?

True. But the sad reality is that not only owners of 4070TIS but even the owners of 7900XT and 7900XTX have no decent upgrade path this gen from neither AMD nor Nvidia. It's like you either go super high end or de facto mid range. The only difference is that Nvidia wants you to pay top dolars for higher midrange (5080) nowdays.

 

[Daven]

Here's your answer: "RTX A5000, which are connected using a 384-bit memory interface" NVIDIA RTX A3000 ; Memory Bus Width, 192 Bit" It looks like all these buses went to enterprise market this time around. Gamers are getting only leftovers nowdays.

Looks like you have to go back almost 15 years for AMD and 20 years for Nvidia gaming GPUs on 64, 128, 256 and 512 only. But you are right. Workstation versions still have the ‘in-between’ busses.

I wonder, is UDNA planned to be chiplet based ? because if they do something like Ryzen with a common IO die across SKUs, we could see larger buses if the economics of doing so are interesting enough for them to go with that... The "golden standard" is essentially the 192-256 bit bracket but honestly, just like with PCIe lanes on CPUs, I wouldn't mind if the amount we got was bumped up to 384-512...

Chiplets would be cool based on the following:

TSMC N2
32 CU
128-bit
8 GB
2.5 GHz
75 W

But four of those together on one package and you would have a monster GPU.

 

[TheGeekn°72]

True. But the sad reality is that not only owners of 4070TIS but even the owners of 7900XT and 7900XTX have no decent upgrade path this gen from neither AMD nor Nvidia. It's like you either go super high end or de facto mid range. The only difference is that Nvidia wants you to pay top dolars for higher midrange (5080) nowdays.

I have a question for you : do you/they need to upgrade ? are the 4070TiS/4080S/7900XT(X) so lacking in the performance department that you have to renew your hardware now ?
my 9070XT doesn't beat the raster of a 7900XTX/4080S and that's okay, I never expected it to do so, I just wanted the newer features, hell if I didn't want the newer features, I'd have gone for an XTX instead, but I see absolutely no reason for anyone who got a PC with parts made in the last 2 generations prior to this one to upgrade now, RTX30/RX6000 is still more than capable to run games very, very well, RTX40/RX7000 are still absolute beasts, why would anyone bother to upgrade ? RTX50 is just RTX40+AI, RX9000 is an actual uplift but if you have the raster, do you need FSR4 ?

I got a 9070XT because I was on laptops for the past decade since I started using PCs (and I'm playing VR and multistreaming to twitch and youtube so AV1 comes in clutch to not put too much strain on the H264 encoder), I don't expect to have to change GPUs for the next 5-7 years unless something goes dramatically wrong in the game industry and they start requiring over 16GB for 1440p gaming or I get the means to get fancier tech.

Chiplets would be cool based on the following:

TSMC N2
32 CU
128-bit
8 GB
2.5 GHz
75 W

But four of those together on one package and you would have a monster GPU.

goddamn, stop, i can only get so erect ! :troll:
even with a chiplet design, couldn't they get higher clocks ? Zen6 is allegedly aiming for high 5/>6GHz and RDNA4 breaks the 3GHz barrier on GPU
I hope they go for something like that that actually becomes capable of challenging the 5090/6090 and put more funding in ROCm, that would finally be enough of an argument to get people to take them seriously enough to steer them away from Nvidia

 

[JustBenching]

Cyberpunk with a 4K texture pack, 24GB is a challenge with that mod.. many, many other games with mods.

He bought a 16gb card and has to defend it to death. No point arguing, pat him in the back and tell him he is okay.

 

[Daven]

I have a question for you : do you/they need to upgrade ? are the 4070TiS/4080S/7900XT(X) so lacking in the performance department that you have to renew your hardware now ?
my 9070XT doesn't beat the raster of a 7900XTX/4080S and that's okay, I never expected it to do so, I just wanted the newer features, hell if I didn't want the newer features, I'd have gone for an XTX instead, but I see absolutely no reason for anyone who got a PC with parts made in the last 2 generations prior to this one to upgrade now, RTX30/RX6000 is still more than capable to run games very, very well, RTX40/RX7000 are still absolute beasts, why would anyone bother to upgrade ? RTX50 is just RTX40+AI, RX9000 is an actual uplift but if you have the raster, do you need FSR4 ?

I got a 9070XT because I was on laptops for the past decade since I started using PCs (and I'm playing VR and multistreaming to twitch and youtube so AV1 comes in clutch to not put too much strain on the H264 encoder), I don't expect to have to change GPUs for the next 5-7 years unless something goes dramatically wrong in the game industry and they start requiring over 16GB for 1440p gaming or I get the means to get fancier tech.


goddamn, stop, i can only get so erect ! :troll:
even with a chiplet design, couldn't they get higher clocks ? Zen6 is allegedly aiming for high 5/>6GHz and RDNA4 breaks the 3GHz barrier on GPU
I hope they go for something like that that actually becomes capable of challenging the 5090/6090 and put more funding in ROCm, that would finally be enough of an argument to get people to take them seriously enough to steer them away from Nvidia

It would be cool if each unit could be powered by the 75W of the PCIe slot. The better process node helps but lower clocks might also be necessary. Right now 32 CUs need 150W at 3.1 GHz on TSMC N4.

Just imagine four 75W chiplets with 32 CUs each. The resulting 128 CU package would be 300W and faster than the 5090!! Getting back to more down to earth power draws is what really gets me going.

 

[RedelZaVedno]

I have a question for you : do you/they need to upgrade ?

In my particular case yes. I've fallen into the rabbit hole of VR flight sim gaming and playing Epic 4/5 pancake games in VR using Praydog's UEVR injector. My PC 5700X3D/4070TIS is just not good enough for this. I'd basically need 4090 level of GPU and 7800X3D or better 9800X3D CPU performance to get to desired stable 60fps at 8-10K resolution using DLSS upscaling. That's why I don't know what to do. 5080 feels kind of weak upgrade over 4070TIS, especially 256-bit bus limitation at these resolutions given the price and 5090 is just too stupidly expensive.

 

[JustBenching]

I wonder if it displaces / replaces 5080 at same price. Surely that would make a few people happy, not the people that come to TPU to pay out on Nvidia of course, I mean actual buyers of their cards Wether they had 'enough' VRAM or not.

Think about it. People are complaining that nvidia doesnt offer more than 16gb of vram in a thread regarding nvidia offering 24GB of vram. All the while pretending they are the enlightened ones for rocking an 800$ midrange mediocre gpu with... 16gb of vram. You cant make this crap up

 

[Chrispy_]

Adding more VRAM to a card that already has enough makes no real sense. More VRAM is always welcome, but from a future-proofing perspective, 5080 buyers are likely always chasing the next top-end GPU. They'll probably upgrade their 5080 after a couple of years, so by the time games that need 24GB VRAM come out, they won't even have the 5080 any more and it won't be fast enough by then to run AAA games at max settings/4K60+ or 144p120+

The 5080 Super needs more cores because it's basically capped out on compute and there are already factory overclocked variants of the regular 5080 that offer what the Super promises.

The 5070S is going to be hotly anticipated because the vanilla 5070 is a good GPU that has a questionable amount of VRAM for it's 1440p/4K capabilities. Giving it 18GB GDDR7 solves its only real drawback, and if the core count is increased even a tiny bit then that and the 18GB would make it a great option.

 

[TheGeekn°72]

In my particular case yes. I've fallen into the rabbit hole of VR flight sim gaming and playing Epic 4/5 pancake games in VR using Praydog's UEVR injector. My PC 5700X3D/4070TIS is just not good enough for this. I'd basically need 4090 level of GPU and 7800X3D or better 9800X3D CPU performance to get to desired stable 60fps at 8-10K resolution using DLSS upscaling. That's why I don't know what to do. 5080 feels kind of weak upgrade over 4070TIS, especially 256-bit bus limitation at these resolutions given the price and 5090 is just too stupidly expensive.

then get a 4090 or a 7900XT(X) ? you don't have to buy the latest gen on the market since there are alternatives
a 4090 might be hard to get for a decent enough price but I'm sure there's still 7900XT(X) still available out there and those aged really well on top of having >16GB of VRAM
to be honest, I think you might get a more significant boost from switching to AM5/DDR5 (a 9700X/7800X3D with faster memory should still be a significant enough upgrade that you shouldn't have to shell out the cash for a 9800X3D -unless you find a good deal or have the cash for it that is-) than getting a new GPU but flight sim ARE graphics hungry after all so...

 

[wolf]

It was supposed/estimated by users............... Just like RTX 5070 was supposed to have 16 GB VRAM.

So supposed to by a small number of enthusiasts with high and often disproportionate demands.

The main reason I don't buy class 80 cards from Nvidia is absurdly low VRAM capacity. They have destroyed this class of cards. Complete nonsense.

6800XT had 16GB of VRAM in 2020 already. It is taking Nvidia 5 long years to decide to offer more than 16GB on class 80 cards. It feels like consumers need to beg for extra VRAM on a product in $1,000 price class. That's why I showed them a middle finger long ago.

My 3080 kicked frikken butt for 4+ years solid at 3440x1440 and then a 4k120 OLED, and still performs incredibly well in my sons rig on a 4k120 display. Between a 3080 and a 6800XT, knowing what I know now and buying either new when new I'd take the 3080 10 times out of 10. She's is ageing like a fine wine with the way the feature set and outright performance in contemporary titles has matured. No doubt in my mind that was worth it over just more VRAM and ongoing shortcomings in every other way, but you do you, show them that middle finger I guess

 

[RedelZaVedno]

then get a 4090 or a 7900XT(X) ? you don't have to buy the latest gen on the market since there are alternatives
a 4090 might be hard to get for a decent enough price but I'm sure there's still 7900XT(X) still available out there and those aged really well on top of having >16GB of VRAM
to be honest, I think you might get a more significant boost from switching to AM5/DDR5 (a 9700X/7800X3D with faster memory should still be a significant enough upgrade that you shouldn't have to shell out the cash for a 9800X3D -unless you find a good deal or have the cash for it that is-) than getting a new GPU but flight sim ARE graphics hungry after all so...

Cheapest 2nd hand 4090 I could find costed 1800€ here in Europe so I might as well get new 5090 for 2300€ if I choose to follow this stupidly expensive route. As for 7900XTX, it is hardly an upgrade over 4070TIS. True you get 15-20% raw rasterisation performance uplift out of it, but you lose DLSS4 support which is a game changer when it comes to VR as even 4090/5090 need it to run most pancake games in VR via UEVR injector smoothly.

 

[Hereticbar]

Show me where 16GB of VRAM is not enough for 1440p gaming. Better GPU horsepower will be way, way better improvement over more VRAM and by far!!!

I would rather he show me next weekend’s lottery winning numbers! I mean he seems to be pretty “good” with his predictions. And I would buy everyone in this thread a 5090 when I win the lottery, then we wouldn’t have to argue if 16GB is enough or not.

Don’t expect to be able to play games at 5K though, because according to his very “accurate” predictions we will need 48-64GB of VRAM….

 

[scooze]

Very good super video cards. A lot of crazy people without understanding why they need additional memory will hunt for new versions and create the ground for very good prices on basic versions while they remain in stock.

 

[Krit]

Adding more VRAM to a card that already has enough makes no real sense. More VRAM is always welcome, but from a future-proofing perspective, 5080 buyers are likely always chasing the next top-end GPU. They'll probably upgrade their 5080 after a couple of years, so by the time games that need 24GB VRAM come out, they won't even have the 5080 any more and it won't be fast enough by then to run AAA games at max settings/4K60+ or 144p120+

The 5080 Super needs more cores because it's basically capped out on compute and there are already factory overclocked variants of the regular 5080 that offer what the Super promises.

The 5070S is going to be hotly anticipated because the vanilla 5070 is a good GPU that has a questionable amount of VRAM for it's 1440p/4K capabilities. Giving it 18GB GDDR7 solves its only real drawback, and if the core count is increased even a tiny bit then that and the 18GB would make it a great option.

Many people are not getting it i also upgrade almost every gen. If you are PC enthusiast you will do it anyways because of curiosity alone. Now imagine gpu that has RTX 4090 gpu power but only 16GB of vram and now imagine RTX 5080 with 24GB of vram but the same speed in terms of compute. Now imagine test in 200 games at 1440p and 4k almost in every case 16GB RTX 4090 will kick the shit out of that 24GB RTX 5080 it will not be even close comparision on average at all. GPU power is more important than vram amount!

 

[Dyatlov A]

That's why i'm saying that vram is enough but gpu itself it too weak.

Agree, the 5080 series are too weak, and more memory won’t help.

 

[LittleBro]

So supposed to by a small number of enthusiasts with high and often disproportionate demands.

If by that you mean people who actually expect some progress and added value to make up for increased costs, then yes.

Expecting 16 GB VRAM for $549 RTX 5070 was definitely not something bold or out of this planet.
3 predecessors from RTX 4000 series have same amount, while 2 lower tier GPUs (4060 Ti, 5060 Ti) have even more.

Many people are not getting it i also upgrade almost every gen. If you are PC enthusiast you will do it anyways because of curiosity alone. Now imagine gpu that has RTX 4090 gpu power but only 16GB of vram and now imagine RTX 5080 with 24GB of vram but the same speed in terms of compute. Now imagine test in 200 games at 1440p and 4k almost in every case 16GB RTX 4090 will kick the shit out of that 24GB RTX 5080 it will not be even close comparision on average at all. GPU power is more important than vram amount!

That's true but with raising demand for AI workloads, RTX paired with 24 GB memory offers interesting value in 1000-1100 € territory for these purposes.

 

[Teiji]

Any chance they will upgrade the fire hazard cable? I really really want to upgrade my 3080 to 5080 Super but not if it still has the same dumpster fire cable of current gen.

 

[JustBenching]

Agree, the 5080 series are too weak, and more memory won’t help.

The 3rd fastest gpu on the planet Earth is too weak.

Any chance they will upgrade the fire hazard cable?

It has been upgraded since 2022 with ADA. No more 8pin fire hazards, all nvidia gpus use the 12vh now, no worries man.

 

[Krit]

The 3rd fastest gpu on the planet Earth is too weak.

You can't see beyond your nose. Only 15% faster than way overpriced higher midrange RTX 4080 that's a very slow gpu gen to gen. RTX 5080 Super will be another slow gpu but with high price as always since blackwell. Basically you are paying 1200$ for peanut leftovers not very smart thing to do.

 

[terreror]

The main reason I don't buy class 80 cards from Nvidia is absurdly low VRAM capacity. They have destroyed this class of cards. Complete nonsense.

6800XT had 16GB of VRAM in 2020 already. It is taking Nvidia 5 long years to decide to offer more than 16GB on class 80 cards. It feels like consumers need to beg for extra VRAM on a product in $1,000 price class. That's why I showed them a middle finger long ago.

An equivalent of class 80 card from AMD, 7900XTX, offered 24GB of VRAM in 2022 already. So, not on a refresh, but from day one. As 5080 Super is already maxed out on die, the only thing they can offer is upgraded VRAM modules from 16Gb to 24Gb, aka 3GB. I hope they dont charge two kidneys for 8GB extra VRAM.

And yet..the 6800xt (and even the 6900xt) are overall slower than a 3080 10gb, 5 years later. Even in vram heavy titles.. 5080 is NOT a 4k GPU, the situation will be the same as with the 3080 - the core will bend the knee before it hits the VRAM limit. And when it does - DLSS will keep the VRAM usage at bay. Mentioning 6800xt and 7900xtx as an example isn't giving you any credit, either. While a 3080 10gb is perfectly cabale of 60fps in newer titles at 1440p with DLSS.., 6800xt isn't. FSR 3 is just unusable while the gpu is not up to the task at native. Next year is 7900xtx's turn..

In addition - nvidia and amd handle textures and vram differently. AMD cards usually need more vram at the same settings than an equivalent nvidia gpu.

On the topic - if the core remains the same and the only difference is the vram and clocks - 5080 Super provides nothing more and will last just as long as the vanila 5080. Playing a game with 15 fps instead of 5 makes no difference..

 

[Hecate91]

It's called pay more to get less.

The more you buy, the more Nvidia saves.

If by that you mean people who actually expect some progress and added value to make up for increased costs, then yes.

Expecting 16 GB VRAM for $549 RTX 5070 was definitely not something bold or out of this planet.
3 predecessors from RTX 4000 series have same amount, while 2 lower tier GPUs (4060 Ti, 5060 Ti) have even more.

The people who will buy Nvidia cards regardless of progress or value probably have very little concern for either.
I expected more than 12gb of vram from the 5070, cards such as the 5060 8gb also should have more. On a card like the 5080 it doesn't make any sense to add more as vram isn't the limitation.

That's true but with raising demand for AI workloads, RTX paired with 24 GB memory offers interesting value in 1000-1100 € territory for these purposes.

AI is likely the reason for 24gb, they aren't doing gamers any favor as some here believe. The x80 sku should be $800 euro/USD not over 1,000 but with the AI marketing it'll probably be 1,100.

Any chance they will upgrade the fire hazard cable? I really really want to upgrade my 3080 to 5080 Super but not if it still has the same dumpster fire cable of current gen.

The 5080 and 5090 should've had 2x 12vhpwr plugs but it seems like Nvidia wanted to save a few cents on components over having a safe and reliable power connector.

 

[JustBenching]

You can't see beyond your nose. Only 15% faster than way overpriced higher midrange RTX 4080 that's a very slow gpu gen to gen. RTX 5080 Super will be another slow gpu but with high price as always since blackwell. Basically you are paying 1200$ for peanut leftovers not very smart thing to do.

I know, the 3rd fastest gpu is very slow. Buying it is not smart, on the other hand buying a 7900xt, swapping it for a 7800xt and then swapping it for a 9070xt is a high IQ move, it punches deep into Jensen's capitalism and consumerism. Bravo.

My 3080 kicked frikken butt for 4+ years solid at 3440x1440 and then a 4k120 OLED, and still performs incredibly well in my sons rig on a 4k120 display. Between a 3080 and a 6800XT, knowing what I know now and buying either new when new I'd take the 3080 10 times out of 10. She's is ageing like a fine wine with the way the feature set and outright performance in contemporary titles has matured. No doubt in my mind that was worth it over just more VRAM and ongoing shortcomings in every other way, but you do you, show them that middle finger I guess

Yeah DLSS Transformer made it a kickass card actually. It will get driver support for many more years and couple that with all the DLSS advancements that it will be getting, it will age better than rdna 3, let alone 2.

 

[Krit]

I know, the 3rd fastest gpu is very slow. Buying it is not smart, on the other hand buying a 7900xt, swapping it for a 7800xt and then swapping it for a 9070xt is a high IQ move, it punches deep into Jensen's capitalism and consumerism. Bravo.

All or nothing! nvidia is the leading company that regulates everything and right now it's gamers or average pc users enemy no.1 becouse they will not stop shrinkflation and prices will not change or even worse will increase. AMD also is doing nothing just following nvidia steps and copying its behavior look at RTX 5060 8GB vs RX 9060 XT 8GB how do you think amd will sell that shit?

 

[JustBenching]

All or nothing! nvidia is the leading company that regulates everything and right now it's gamers or average pc users enemy no.1 becouse they will not stop shrinkflation and prices will not change or even worse will increase. AMD also is doing nothing just following nvidia steps and copying its behavior look at RTX 5060 8GB vs RX 9060 XT 8GB how do you think amd will sell that shit?

Nvidia doesn't regulate anything. Nvidia CANNOT stop Intel or AMD from launching a GPU that roflstomps the 5090 in performance for 500$.