According to Videocardz's sources, NVIDIA is now planning to launch the RTX 3080 20 GB variant sometime in December. We had first heard about the 20GB variant before the Ampere series was announced and it had originally been planned for release just after AMD's Big Navi. This new timeline would represent a delay of about a month or so - assuming AMD goes with a hard launch of their Big Navi series and NVIDIA would not have made such a decision lightly.
 
https://wccftech.com/nvid...or-launch-in-december/

Told ya, I called it😼
An exciting 300 series card with more than 12GB of RAM in place😺.
That also means AMD will be releasing GPUs with the same amount of RAM😎👍

aka_STEVE_b
I have come to believe that Nvidia never actually planned to mass produce their own 3080 FE card in any kind of decent amounts on purpose.... until they saw what AMD has.
 They just wanted the early reviewers hype and news about their awesomeness....to get people in a frenzy.
 
 

It's called Marketing strategy. Nvidia is a good on feeding hype

Brad_Hawthorne
Not sure I'd understand a 12GB 3900 though.

It's not an 8K card.

Cut the VRAM in half for the huge majority of users who don't need 24 GB, and you can cut costs. Also, VRAM with half the density will easily have much better latencies, therefore better throughtput and better performance.
Win, win. It would be very marketable for the majority of potential buyers. A cheaper, faster gaming card. Also better world record potential. It would be better for everyone other than high class production (a small majority anyway).

A 20 GB 3080 doesn't seem very marketable. More expensive. Worse latencies. Slightly lower performance for the majority of users who don't actually ever use more than 10 GB of VRAM (12 to 14 GB actually available on standard 10 GB 3080 due to built-in hardware compression/decompression). And so forth.

It will be very interesting to see how they market a 20 GB 3080. I think it will prove to be pointless and honest reviews will reveal as much. Double VRAM density cards have always proven to be worse for the majority of users.

Dont see a point to spend what is probably going to be extra 300$ for no extra performance, you will be upgrading way before vram becomes an issue, and from todays amd event those fps number put big navi a bit slower than 3080 while big navi does have the 16gb of vram 

Really should've been a 16GB base chip, not 10.

Xavier Zepherious
so it begs to differ - we could most certainly use the extra ram if games are currently going to be designed to need it 

Like I said, we will see, won't we?
If a double VRAM card finally proves to perform better for the majority, it will be a first.  ;)
It's a matter of fact that if you double the VRAM density, the VRAM has lower performance; and a matter of fact that the GPU core performance has always been the limiting factor at high resolutions whenever these comparisons have been made in the past.
Have the tables finally turned? No need to argue. We will see.

Personally, a higher performance, cheaper 3090 makes more sense to me; versus a lower performance, more expensive 3080. A "12 GB" 3090 could store up to around 16 GB of texture data utilizing its built-in hardware VRAM compression/decompression. That should be more than enough for 4K gaming for a long time to come. 8K gaming? What are you talking about? The card can't handle native 8K gaming. And if you use DLSS to achieve a playable fake 8K resolution, you have nothing to worry about VRAM-wise.

Again, with the same core, a 20 GB 3080 WILL BE SLOWER than a 10 GB 3080 UNTIL the 10 GB 3080 runs out of VRAM. That's crazy, but true. So, like I said, we will have to wait and see if/when the 10 GB (12-14 GB usable) 3080 runs out of VRAM. Those reviews will come, and they will be interesting. It's just crazy to market something like 'this card costs more and will be slower, until it catches its stride and starts to shine right before the core runs out of ability and the FPS drops to a point that it no longer even matters'. The same thing which has been repeated for a decade.

The performance would be less at first, and then there would be some window where the performance would be better, before the core falls on its face and it doesn't even matter any more. How big is that window going to be? That's the question. In the past, that window has been so small that it hasn't made sense for the majority of users.

well 20GB...  ok...
 
16GB i would get it... but 20GB? that's a LOT considering 3 years ago ( in 2017) 11GB was already a ton of Vram on the 1080's Ti!
 
am i missing something here? are games gonna be that Vram needy next 4 years??
 
fine then .... as long as the performance is there i guess no one cares how much Vram they stuff in there... just for show i guess!!
 

8K is still a ways off
 
4K is sill very extreme
 
 

CraptacularOne
It's probably going to get a few more SMs unlocked and slot in between the 3090 and 3080 in price and performance. There is a giant gap left between $699 and $1499 that I'm sure Nvidia is eager to fill. 

The problem is that there isn't room between the 3080 and 3090 in terms of performance. I'm not arguing business strategy, but the 3090 (which is not a TITAN, period) is bad enough--you have to admit that an 8% performance increase would be an utter joke of the x80 Ti branding at any price. If NVIDIA asks two or three hundred dollars more for that, we've regressed all the way back to the days of the 8800 Ultra, lol.
 
Alright, there would be a lot more VRAM, but I still haven't gotten used to these prices.
 
Regarding VRAM, it seems to me that some people are basing their assumptions on the progress of the last seven years, and not taking into account the fact that VRAM usage is going to increase significantly overnight when the first next-gen-only games are released. My understanding is that that is not supposed to happen until two years after the consoles release, but I don't think it's at all unlikely that it could occur sooner.
 
With the massive price increase of GPUs, I would assume there are many people who are trying to keep their high-end GPUs longer than they traditionally would. I know I am. If I buy a 30-Series card six months from now, I expect to use it for at least 2 years. I certainly wouldn't want to be using a 10GB card 4 years after I bought an 11GB card, and 5.5 years after the first 11GB xTi card, in a new generation of console-to-PC ports.

ty_ger07
The RTX 3xxx series has built in hardware VRAM compression and decompression and NVIDIA advertises that the 3080 should be able to store 12 to 14 GB of data in its 10 GB of VRAM.

I already said all of that at least once before.

Yes, and I passed over it because I didn't necessarily agree with your assumptions. I'll be glad to accept it if you can provide the info you're working from, but the only new compression tech I'm aware of accelerates transfer rates between storage and VRAM. This could potentially reduce required VRAM by virtue of the ability to stream data in fast enough to avoid keeping unneeded data on the card.
 
But, this won't "effectively increase" VRAM at all, because the Series X uses the same compression. I.e. the PC has to use it just to keep up with the console--and then you factor in the standard significantly higher VRAM usage over consoles due to the complete lack of PC-focused coding and optimization.
 
If you are talking about something else, I'll take a look at it, but this is the only thing I'm familiar with.
 
Even if next-gen games have to be designed to the level of the Series S and not the Series X, there's still a huge jump in shared memory amount, so if games are already utilizing 7 or 8GB at 4K max, I don't assume that 10GB will cut it for next gen.
 

ty_ger07
The 3090 is not an 8K capable card, unless DLSS is used.

I'm glad other people are saying this too, lol. I find it mind-boggling that people are actually talking about "8K gaming" like the phrase makes any sense, just because it fell from the lips of Jen-Hsun once or twice. Something like 5% of gamers are on 4K right now. Nobody games on an 8K display. Nobody is going to be gaming on an 8K display before the 30 Series is very old news. And no, supersampling doesn't count: 1920x1080 at 4x DSR isn't "4K gaming".

ty_ger07
I'm not talking about nvcache. I am talking about tensor memory compression. NVIDIA says that it compresses textures in vram 20-40% effectively making it as if you have 20-40% more vram. It seems like there aren't any examples of it being implemented yet though, for some reason.
I guess time will tell.
Maybe I'm mistaken. Maybe it doesn't actually exist. Maybe it is just a CUDA thing.

The thing is, unless Nivida makes it a driver level implementation, it's not going to get used by game devs in engines unless Nvidia pays the company to implement it. The problem with Nvidia proprietary features is they don't get used unless Nvidia appeals the the pocketbook of the game dev. Nvidia throws marketing money at stuff like that for presentation demo content. There is always resistance from implementing it without a reason though. I get the feeling game devs feel a bit of resentment being made into the marketing arm for a hardware manufacturer unless they're getting a good chunk of cash for it. It's just like multi-GPU scaling in DX12. Microsoft has it there for devs to implement but it never will be used because it's not just there transparently at a driver level. If a game dev has to invest time and money to implement something there needs to be a ROI for it or you'll never see it done.

I have noticed that several recent games are saturating the 8GB of VRM I have on my RTX 2080 suggesting its time to move to 16GB
 
Along comes Ampere and more of the same, going cheap on the VRAM
 
AMD has not been so cheap and even my RTX 480 has 8GB and that card competed with 3GB and 6GB cards
 

I agree, 4K nowhere near normalized.