Monday, May 11th 2020

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

PC hardware focused YouTube channel Moore's Law is Dead published a juicy tech-spec reveal of NVIDIA's next-generation "Ampere" based flagship consumer graphics card, the GeForce RTX 3080 Ti, citing correspondence with sources within NVIDIA. The report talks of big changes to NVIDIA's Founders Edition (reference) board design, as well as what's on the silicon. To begin with, the RTX 3080 Ti reference-design card features a triple-fan cooling solution unlike the RTX 20-series. This cooler is reportedly quieter than the RTX 2080 Ti FE cooling solution. The card pulls power from a pair of 8-pin PCIe power connectors. Display outputs include three DP, and one each of HDMI and VirtualLink USB-C. The source confirms that "Ampere" will implement PCI-Express gen 4.0 x16 host interface.

With "Ampere," NVIDIA is developing three tiers of high-end GPUs, with the "GA102" leading the pack and succeeding the "TU102," the "GA104" holding the upper-performance segment and succeeding today's "TU104," but a new silicon between the two, codenamed "GA103," with no predecessor from the current-generation. The "GA102" reportedly features 5,376 "Ampere" CUDA cores (up to 10% higher IPC than "Turing"). The silicon also taps into the rumored 7 nm-class silicon fabrication node to dial up GPU clock speeds well above 2.20 GHz even for the "GA102." Smaller chips in the series can boost beyond 2.50 GHz, according to the report. Even with the "GA102" being slightly cut-down for the RTX 3080 Ti, the silicon could end up with FP32 compute performance in excess of 21 TFLOPs. The card uses faster 18 Gbps GDDR6 memory, ending up with 863 GB/s of memory bandwidth that's 40% higher than that of the RTX 2080 Ti (if the memory bus width ends up 384-bit). Below are screengrabs from the Moore's Law is Dead video presentation, and not NVIDIA slides.
As for performance, the "GA102" based prototype is allegedly clocking 40 percent higher performance than the RTX 2080 Ti at 4K UHD resolution in poorly optimized games, 50% higher performance on optimized games, and up to 70 percent performance in the "best case scenario" (a game that's been optimized for the "Ampere" architecture). We know from older leaks that by increasing the number of streaming multiprocessors, NVIDIA is doubling the CUDA core : RT core ratio compared to Turing, resulting in more RT cores per tier; and increased ray-tracing performance.

Each "Ampere" RT core is able to process 4x more intersections per unit clock-speed than "Turing." The tensor core count is also reportedly going to see an increase. The focus on ray-tracing and AI performance increase could give game developers the freedom to cram in more RTX effects per title, letting users disable what they want on older "Turing" cards. Performance limitations on "Turing" made developers choose from the RTX feature-set on what to implement. With "Ampere," NVIDIA could introduce DLSS 3.0, an updated image quality and performance enhancement. NVIDIA could resurrect a hybrid memory technology similar to AMD's HBCC, called NVCache, which spreads video memory across the video memory, the system memory, and flash-based storage.
Lastly, there's more clarity as to what silicon fabrication process NVIDIA could use. Apparently, NVIDIA will spread its product stack between two kinds of 7 nm-class nodes. The higher-end ASICs, such as the "GA102" and "GA103," could be built on 7 nm EUV nodes such as the TSMC N7+; while the smaller ASICs could be built on conventional DUV-based 7 nm-class nodes such as the N7P or even N7.

Don't pull your wallets out just yet. The launch schedule points to May 2020 (GTC) being focused on HPC parts based on "Ampere," such as the Tesla A100 and DGX A100 system.

In September 2020, NVIDIA will hold a separate event specifically to launch the next-generation GeForce, very close to "Cyberpunk 2077" release.
Source: Moore's Law is Dead (YouTube)
Add your own comment

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

#1
dyonoctis
Looks good on paper...RDNA2 might have to be insanely good if we want competition to be healthy again.
"new control panel merged with geforce experience" Ho boy. Some people are going to be so mad about that.
Posted on Reply
#2
BArms
Finally HDMI 2.1 support. My current 4K TV is a 120hz panel but there's no way to feed it a 4K signal @ 120hz, and that was from 2018.
Posted on Reply
#3
AsRock
TPU addict
And are nVidia keeping the best binned chips for them self's again too ?, sounds like a great way to screw the company's that sell your silicon.
Posted on Reply
#5
gmn17
2 sleeps to announcement can’t wait
Posted on Reply
#6
Camm
There was a lot of contradictory information in MooresLawIsDead's video (the big one being DLSS 3.0 info). I'm filing this under FUD until we get some form of announcement from Nvidia.
Posted on Reply
#7
xkm1948
Im definitely getting this. Increased Tensor performance for consumer tier pricing is a huge bonus
Posted on Reply
#8
renz496
So far he is the only one getting this rumor from "insider source". I'm not hearing others talking about this.
Posted on Reply
#10
InVasMani
Looks good at a glance and I think it's safe to say that pretty much everybody presumes it'll be one of the better generational Nvidia upgrades that we've seen in a few generations. It appears as if they've got a basic layout plan in order from this last RTX generation and are shrinking it down further to 7nm while refining it to be faster and more flexible. It seems like in terms of ray tracing details it'll be x4 the performance to work with from the RT cores alone which is good. The part about the global illumination, shadows, and reflections though is interesting I'd take that with a grain of salt perhaps in relation to the x4 RT cores processing it might allude to it being able to utilize that x4+ RT processing power and diveying it up more flexibly. I presume they might say it's got x4 more RT performance to work with they can divide it up in a variable rate mannerism between those two subsets of features global illumination, shadows, and reflections in whichever manner a developers feels is the most appropriate use of hardware. Regardless it still sounds like good progress. The NVCache response to HBCC is interesting and sounds a awful lot like AMD's StoreMi actually coincidentally or like PrimoCache. To me it seems like they may take their conventional disk based shader cache and place it into a SSD/RAM accelerated tiered cache of sorts.

Still really interested in RDNA2 as well it sounded a lot like AMD was going to do sort of a single PCB dual chip GPU and make it more of a standardized GPU for the entire generation of GPU's? Unless I interpreted what I had read in some article on it wrong I only read one article that hinted toward that kind of thing, but I could be wrong about that I haven't kept close tabs on it overall other than some tid bits on about RNDA2 to do with consoles. I'd expect more from Nvidia this generation than AMD for GPU's personally given they are doing a die shrink, but who knows. It's a interesting time for GPU's especially with all the flexibility of the added PCIe 4.0 bandwidth to work with.
Posted on Reply
#11
wolf
Better Than Native
To be taken with appropriate pinch of salt, however if these statements are true
NVIDIA is doubling the CUDA core : RT core ratio compared to Turing, resulting in more RT cores per tier; and increased ray-tracing performance. Each "Ampere" RT core is able to process 4x more intersections per unit clock-speed than "Turing."
That will be pretty darn great for just about any teir of card. I could see a 3060 or above potentially being a compelling upgrade for a GTX1080 sort of class card if if it can out-muscle it to the tune of 30%+ in rasterization then have likely 2080Ti++ level of RT hardware/capability on board. Sprinkle on DLSS 3.0 in upcoming games (assuming the universality of it is realized) and man am I tempted.
Posted on Reply
#12
M2B
My biggest concern is the amount of VRAM we'll be getting.
The 10 & 12 GB cards (3080/3080Ti) are going to be fine but if midrange cards such as the 3070 only get 8GB it's going to be disappointing.
I don't think 8GB is going to cut it long-term.
I hope nvidia gives more options with more VRAM, I would pay an extra 100$ for the 16GB Version of the 70 class Card.

4x better RT performance doesn't necessarily mean 4x higher framerates when RT is enabled, it could mean 4x faster HW acceleration which at the end should give different results depending on the workload, so don't expect some magic to happen. The Performance cost of Ray Tracing is not going anywhere, it could be reduced but not completely eliminated.
Posted on Reply
#13
Tsukiyomi91
Imagine if there's an RTX3060 (low end Ampere) making the 2080Ti a run for its money...
Posted on Reply
#14
The Quim Reaper
Tsukiyomi91Imagine if there's an RTX3060 (low end Ampere) making the 2080Ti a run for its money...
Maybe in RT performance, but it will still be well behind in traditional raster graphics.

..and if it does turn out to be the case, It says more about how poor the first generation of RT cards are/were, than it says about how good Ampere might be.
Posted on Reply
#15
T3RM1N4L D0GM4
RTX3060 seems like a decent upgrade from my actually GTX760 and my 3440x1440 screen.... :respect:
Posted on Reply
#16
ZoneDymo
Decryptor009AMD needs a miracle.
I think its more a case of the consumer/the world needs to have AMD get a miracle.
We all know what Intel did....or rather did not do due to lack of competition.

ALso I know it does not matter but I do find it kinda boring the most high end unit in this segment not using HBM memory.
T3RM1N4L D0GM4RTX3060 seems like a decent upgrade from my actually GTX760 and my 3440x1440 screen.... :respect:
yeah, but probably....bout 500 dollar pls
Posted on Reply
#17
SIGSEGV
not good at all for us as consumers when there's no competition. expect nothing.
Posted on Reply
#18
Vayra86
wolfTo be taken with appropriate pinch of salt, however if these statements are true



That will be pretty darn great for just about any teir of card. I could see a 3060 or above potentially being a compelling upgrade for a GTX1080 sort of class card if if it can out-muscle it to the tune of 30%+ in rasterization then have likely 2080Ti++ level of RT hardware/capability on board. Sprinkle on DLSS 3.0 in upcoming games (assuming the universality of it is realized) and man am I tempted.
That was my thought too, and the main reason I've been sitting on that exact same card now for quite a while... and it still has enough juice to keep going actually.

The 1080 > 3070 jump will already likely be significant, RT is bonus. 1080 > 3080 will be over 50%. I'm also pretty optimistic about the ratio / increase of RT, it was clear something had to give, the balance seems alright. Its also clear they use DLSS to make room for that in terms of performance. Overall I get the impression Nvidia is holding on to their performance increase in rasterized ops per gen. Very nice. This makes RT 'just bonus'.

And this is what I meant when I said 'Just wait' when Turing dropped. Its nice RDNA2 is going to compete. Thanks, AMD.
Posted on Reply
#19
ARF
renz496So far he is the only one getting this rumor from "insider source". I'm not hearing others talking about this.
Most probably fake news.

RTX 2080 Ti doesn't need 3 power connectors.

2 x 8-pin is good for 75 + 2 x 150 watts.
Posted on Reply
#21
Sunny and 75
Interesting progression:

- 4x better Ray Tracing performance
- Global illumination, shadows and reflections all at the same time
- Higher core clocks (Above 2GHz)


September is going to be a proper time to upgrade for sure!
Posted on Reply
#22
Otonel88
When I saw: triple fan design, dual 8 pin connector, I am just thinking about a high TDP card, which is contradicting the whole 7nm design process, unless this is just rumors.
7nm should be more efficient bringing the TDP down, so why the double 8 pin and tripple fan design.
Posted on Reply
#23
Fourstaff
Otonel88When I saw: triple fan design, dual 8 pin connector, I am just thinking about a high TDP card, which is contradicting the whole 7nm design process, unless this is just rumors.
7nm should be more efficient bringing the TDP down, so why the double 8 pin and tripple fan design.
They may be maxing out the power draw to squeeze out every last bit of performance. Products down the stack should be more efficient.
Posted on Reply
#24
djisas
ARFMost probably fake news.

RTX 2080 Ti doesn't need 3 power connectors.

2 x 8-pin is good for 75 + 2 x 150 watts.
2 x 8-pin + pcie = 150 x 2 +75 = 375 max power draw for the card...
Posted on Reply
#25
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.
Posted on Reply
Add your own comment
Dec 2nd, 2024 14:29 CST change timezone

New Forum Posts

Popular Reviews

Controversial News Posts