NVIDIA RTX 3080 Ti and GA102 "Ampere" Specs, Other Juicy Bits Revealed

[bug]

Specs aside, I wonder where the 3060 will be positioned in the stack.
With Turing Nvidia made a mess with 1660 and 2060, so it was pretty hard to tell where the actual mid-range was/is. I don't imagine this separation will be there with Ampere.

 

[ObiFrost]

Yeah, I don't sense it. 4 times the RT performance of 2080Ti in 3060? RT performance all the way through MX550? "GTX is dead", rofl, Nvidia will find a way to squeeze in 17 series one way or another. He also said Ampere Titan should do a decent "8K" gaming when we have no native support for 8K gaming yet (talking about in-game, not HDMI 2.1).

 

[Recus]

Moore's Law is Dead is AdoredTV's brother in lie. Don't believe their lies.

 

[ShurikN]

Everything seems reasonable except the 2.5GHz boost clocks.

If the performance uplift is as stated you can say goodbye to a $300-350 RTX3060

 

[Ravenmaster]

RTX 3080Ti specs just gave me a nerdgasm

Finally I'll be able to go 4k 120hz with my LG C9 OLED via full HDMI 2.1. And with Displayport 2.0, the card is setting up to be quite future proof for even the next gen monitors if i choose to change my display

 

[Vayra86]

I do think the general, big changes are true here. Double RT ratio compared to Turing seems plausible, because it is very clear even a 2080ti is lacking in that respect. The combination of that with an IPC bump and efficiency on VRAM also sounds very Nvidia and logical.

Triple power connector and triple fan... careface. AIBs do what they like, Nvidia is not immune to marketing either. If triple power plugs sell, they will use them, and give a nice little 'oh but its so much more stable now' BS story to go with it. Triple fan... heat... just the fact it is 7/8nm does not mean it runs cooler, and I think there are enough examples of that already.

The gist of it; 30~40% perf bump per price segment/tier, maybe 50% best case which is really quite good and as expected, and that is coming from Turing non Super probably, which means the Supers are not immediately useless which is also Nvidia's MO.

It all fits quite well tbh, bar some insignificant details. And the NVCP+GFE thing... we'll see how that goes. It also means they can't get around revamping the NVCP which in all fairness could do with a little facelift by now. It also mains greater integration of the added value stuff like Ansel, Shadowplay, Freestyle, streaming, etc etc etc. Not necessarily a bad thing.

What's also a big USP is the continued focus on DLSS. If they are indeed approaching it ultimately as a 'fits all' solution, that is one killer feature to enable high res gameplay and has the potential to massively shift the market in several ways; not just for 1440p/4K gaming, but also for VR. That and cheap headsets will drastically lower barrier of entry.

Interesting times ahead...

 

[ZoneDymo]

2 x 8-pin + pcie = 150 x 2 +75 = 375 max power draw for the card...

ermm yeah? that is what they said...

 

[QUANTUMPHYSICS]

I have a EVGA 2080Ti Black.
All Nvidia really needs to do here is blow the 2080Ti out of the water with the 3080Ti.
I'm gonna buy it anyway, I just need a good reason to be excited.

My hopes are that the lower end 3000 cards are all superior to the Super versions of the low end 2000 cards.

For the 1080Ti to outperform the 2060 and 2070 is ridiculous. The 2060 should have been ahead of the 1080Ti in every way.

 

[renz496]

Everything seems reasonable except the 2.5GHz boost clocks.

If the performance uplift is as stated you can say goodbye to a $300-350 RTX3060

lol i was expecting the clock will at least reach 3ghz this time.

 

[ppn]

Ti is almost never the full die. 1-2 sm disabled 5120 cuda running at 2.2 at best. but it absolutely needs 24GB to make sense. I guess 4080 will finally be the card with 16GB. the whole line does things wrong. RTX won't matter regardless of how much better it is, every new iteration of it renders the previous gen obsolete and useless, 1-2 games 50 hours, that you forget. It is just wasting die space for the next generations until 1,4nm at least.

 

[techguymaxc]

People who are knowledgeable about this industry over at Beyond3D forums have pointed out that this compiled list of specs and features make no sense, and that the Youtuber has no background in engineering or the industry.
It’s no different than if someone from this forum made a video based on rumors.

 

[bug]

People who are knowledgeable about this industry over at Beyond3D forums have pointed out that this compiled list of specs and features make no sense, and that the Youtuber has no background in engineering or the industry.
It’s no different than if someone from this forum made a video based on rumors.

There may be parts that don't make sense, but more HP when moving to a new node was all but expected.

 

[Turmania]

AMD is at least a generation and a die shrink away from Nvdia. So Rdna2 can be good but it will be nothing in comparison to ampere.

 

[djisas]

ermm yeah? that is what they said...

My bad, i guess i misread it...

 

[bug]

AMD is at least a generation and a die shrink away from Nvdia. So Rdna2 can be good but it will be nothing in comparison to ampere.

I wouldn't say "nothing". Stability issues and lack of DXR aside, 5700 turned up slightly better than the 2060 and 5700XT is basically a $100 cheaper 2070.

The potentially sad news in this leak is that if Ampere is indeed such a beefed up Turing, we'll see Turing pricing once again

 

[Bruno Vieira]

AMD is cooking RDNA2 for too long, so long that nvidia came with another uArch before it

 

[RH92]

Everything seems reasonable except the 2.5GHz boost clocks.

2.5Ghz is the most reasonable info in there , Nvidia is hitting 2Ghz already on 12nm , so it won't be hard to them to hit 2.5Ghz on 7nm .

 

[B-Real]

1080Ti was 80+% faster than a 980Ti, so a 40 to 70 increase (take 55% in general) compared to the 2080Ti is big, but not that big to be honest. Or it can be enough IF they massively drop price.

Moore's Law Is Dead also says that it could be a bigger deal than Pascal over Maxwell. Given the numbers, that's far from that. He also says that the 3060 can be around the 3080Ti. I think it will be around the 3080 as in previous gens.

 

[Th3pwn3r]

1080Ti was 80+% faster than a 980Ti, so a 40 to 70 increase (take 55% in general) compared to the 2080Ti is not that big to be honest. Or it can be enough IF they massively drop price.

Lol. And don't expect price drops, they have no reason for that. The gap between Nvidia and AMD on high end just became that much bigger.

 

[efikkan]

While I get that there is an information vacuum and a desperate demand for information about Nvidia's upcoming masterpieces, this article is nothing but speculation from an opinionator on YouTube. Speculation is fine when labeled as such, many of us enjoy doing so here in the forums all day, but pretending that the information comes from numerous sources inside Nvidia, who are risking their jobs and potential legal action to supply a random guy, that's more than a bit far fetched. While some of this information may turn out to be close to reality, this entire "leak" is just guesswork.

Ti is almost never the full die. 1-2 sm disabled 5120 cuda running at 2.2 at best. but it absolutely needs 24GB to make sense. I guess 4080 will finally be the card with 16GB. the whole line does things wrong.

I wouldn't mind more more memory if it didn't have a large impact on the price, but the card would need a significant increase in memory bandwidth to go along with the increased capacity, otherwise very few games can use it for something useful.

AMD is at least a generation and a die shrink away from Nvdia. So Rdna2 can be good but it will be nothing in comparison to ampere.

AMD is actually a die shrink ahead of Nvidia, and is still playing catchup with Pascal…

 

[Dante Uchiha]

RDNA2 vs Ampere will be quite interesting. AMD says efficiency gains are up to 50% over RNDA1, this could directly translate into +50% gaming performance OR not.

The 2080 ti is approximately 45~50% faster than rx 5700. The high-end RDNA2 should have twice as many shaders as the Navi10, added to the architectural advances we have a GPU with almost twice the performance of 2080ti. The turning point and biggest limiting factor may be bandwidth... We do not have an ultra-fast memory that accompanies such advances.

2.5Ghz is the most reasonable info in there , Nvidia is hitting 2Ghz already on 12nm , so it won't be hard to them to hit 2.5Ghz on 7nm .

Well, that's not how it works. If it worked this way intel would not have the advantage of higher frequencies with 14nm++ vs. TSMC 7nm. I don't say it's impossible, but I doubt it very much. I mean, there must be a reason nvidia preferred to stay at 12nm FFN...

 

[cueman]

well, thouse infos are as all know ,guesses.

but nvidia ampere has one moust massive performance thing,and its fact, it is 7nm process tech.

for 12nm to 7nm mean alot speed for core and bcoz tensor speed are 4 times faster,its sure, ampere is top of its class.


amd rdna 2,3 gpu?

well,still... that moust important, if amd cant move 5nm process tech, its has no way to up performance anyway than install rdna 2 gpu more hardware, meaning more tmus and rops... and if possile more hertz..
its easy way to do it, but, only if you gpu can handle it loading process without skyhigh heat.

amd has not new tech core,its still running 'old' one, like them flahship radeon vii,rdna 2 will be that gpu child, thats fact.
amd get ,maybe new core tech rdna 3, i say might, but then there is nvidia Hopper....

and, as all can read from net, amd not get this time big handicap for process tech, bcoz Hopper ,rdna 3 and even intel has all same,5nm tech thouse toys.


last i want say that 7nm nvidia ampere rtx 3080 ti are NOT flagship from ampere family,no,it is Titan Ampere... i guess about 10% more performance...

hmm as we all know even amds 'real' flagship 7nm rx 5700 xt is almost more than 40% slower than 12nm rtx 2080 ti, i think amd have plenty work to beat that, and 7nm rtx 3080 ti is totally different challenge.


but, we seen it this year and im sure we get soon, month or so performance leaks from ampere...myself i hope 1st leak coming 7nm ampere rtx 3060 performance...if that gpu performance go close rtx 2080 ti scores, amd is bad situation and will be trouble...and i ecept that 7nm rtx 3060 tdp is under 200W and clear.

anyway,we must remembe,these days fficiency is one of moust important think all hardware,and must be big place when give hardware rewards,
so, if 7nm rtx 3080 ti is about 40% faster than 12nm rtx 2080 ti and doing that under 240w, its amazing results what is very very hard broke.

well, as i say, we must wait for results.
also i wait exited also,intel 1st gpu perfomance.... my guess is between rx 5700xt/rtx 2060 super performance, but moust interesting is, i think efficiency,aka fps/watt... like all incoming gpus!
fact is that intel gpu results it 7nm gpu are going 7nm cpu sector also...


interesting!!

 

[Anymal]

Bring it on! Full tower case Phanteks 719 and 750W gold unit are waiting for this goddess. Impressive!

 

[medi01]

"GA102" reportedly features 5,376 "Ampere" CUDA cores (up to 10% higher IPC than "Turing").

2080Ti was 4352.

So, with "up to 10% higher IPC", could be up to 35% faster and with some smart down-scoring of the older cards in reviews, even "faster".

I doubt clockspeed would improve much.

AMD is actually a die shrink ahead of Nvidia, and is still playing catchup with Pascal…

AMD's N7 RDNA is roughly at perf/transistor levels of Turing.

2080Ti is only about 50% ahead of 5700XT, a fairly small 250mm2 N7 chip.
Rumored big RDNA2 is on a better process and twice the size.

Unlike NV, which had lavish R&D budget for a while, AMD GPU group is just warming up and results have been impressive so far, nobody expected Navi to leap that far ahead of Vega.

The days of AMD absence in the high end GPU market will be over once these cards hit the market.
Ryzen penetration will also make users more likely to buy AMD GPUs.

 

[efikkan]

AMD's N7 RDNA is roughly at perf/transistor levels of Turing.

Not very impressive considering Turing have spent a lot of those transistors on additional features, and still manages better efficiency on an inferior node (unless Navi runs at low clock speeds).

2080Ti is only about 50% ahead of 5700XT, a fairly small 250mm2 N7 chip.
Rumored big RDNA2 is on a better process and twice the size.

Don't forget that Nvidia will have access to any new nodes that AMD does have access to.

RTX 2080 Ti is not very impressive over RTX 2080 (Super) because it's held back by power limits. Imagine "RTX 3080 Ti" unleashed on 7nm, it's not going to be a cakewalk no matter how good RDNA 2 is.