NVIDIA Architecture doesn't linearly scale due to how it works., the same is true for Intel Architecture but they're very close in basic design.
Now as a key point-of-note:
Frequency x CoreCount x 2 (Ops/Cycle) = Floating Point Operations / Second
1582 x 3584 x 2 = 11,339,776 MFLOPs or 11.34 TFLOPs [GTX 1080Ti]
1542 x 4352 x 2 = 13,421,568 MFLOPs or 13.42 TFLOPs [RTX 2080Ti]
This is a +15.5% Peak Performance Difference.
Of course is we compare Base Clock., we get 10.61 TF Vs. 11.75 TF... which is only +9.7% Baseline Performance difference.
Keep in mind that GTX 980Ti to GTX 1080Ti was a 50-55% Performance Difference.
We can only assume how well this will actually Clock... remember that it is possible to coax 2050MHz Boost Frequency out of a GTX 1080Ti AIB.
Yet remember we're talking about a balanced Heat Dissipation Profile., so while 12nm (14nm++) does provide the potential for Frequency Improvement (as showcased by Ryzen 2nd Gen), to achieve this AMD didn't change the Die Layout or Dissipation Profile.
In fact, I'd wager that the Higher (OC) "Founders Editions" that also cost more... are likely to be very close to the Maximum Overclock.
Keep in mind here that while the AIB "Stock" is 1350MHz Base / 1545MHz Boost while 1635MHz FE Boost.
Of course there is an argument to be made here that this doesn't mean anything given the GTX 1080Ti is capable of up to 2000MHz over the (effectively 1600Mhz Stock Boost) … but in this case I'm not so sure, as the Titan V tended to tap out at 1780MHz and bare in mind that these were heavily binned chips; as there was very little variation between Review Frequencies or those on the Futuremark Database.
Remember that the Titan V, has a very similar Boost Clock.
As such I'd be surprised if you can get more than 100-150MHz Stable OC on the RTX Series.
The result being is we have the £1,100 RTX 2080Ti [14.73TF] Vs. £670 GTX 1080Ti [15.62TF] … and here's the thing, if we actually compare the potential performance of the £750 RTX 2080 [10.60TF] and £470 GTX 1080 [8.87TF] or £570 RTX 2070 [7.88TF] Vs. £380 (£420) GTX 1070(Ti) [6.46TF (8.19TF)]
Well it just ends up looking worse and worse.
Especially given *all* of the above 10-Series are capable of hitting 2000MHz, very noticeably improving their performance.
There is absolutely Zero Guarantee (like Volta) that Turing will actually clock too far beyond it's Stock Boost.
How does that stack up against the RX Polaris or Vega (GCN 2.0 Architecture)? Well, given that the RX VEGA 56 is effectively on-par with the GTX 1070Ti in *most* titles and the RX VEGA 64 is effectively on-par with a 1080 with Overclocking again in *most* titles. There will be select titles, such-as those using Epic' Unreal Engine 4 (which is heavily co-developed / produced by NVIDIA today), where sure... AMD Graphics just gets thrashed.
Still in most other Scenarios it holds it's own.
And with these Cards *finally* back to MSRP in *most* Regions., this also means Price-to-Performance they're really a coin-flip decision between the NVIDIA equivalents. This means performance wise they're also going to be relatively competitive with the RTX 20-Series as well., and what's more important to keep in mind is that while the GTX 10-Series struggles (but at least is capable of using) Radeon Rays (unlike RTX, which requires specialist Hardware) it doesn't on AMD Hardware... *ANY* GCN 3rd Gen or Better, which is what most people have.
Navi almost certainly (going by the patents filed) is also going to have notable acceleration not just for Radeon Ray workloads, but Machine Intelligence as well as Brute Force Traditional Pipelines... i.e. we could actually see Navi return Terascale 3 DirectX 11 performance to the GCN Architecture., without compromising the Asynchronous Compute.
NVIDIA might be betting heavily on Real-Time Raytracing., AMD isn't... and frankly this makes more sense, but when you bet on "New" methods; you then have to convince Developers to actually adopt it., which history shows, they don't. Not unless you pay them... A LOT.
I might see about putting my nose to the grindstone with Radeon Rays over the next month or so, see if I can't coax my RX 480 to achieve some comparable results to what NVIDIA RTX is showcasing because frankly I actually believe it's more than possible to achieve some similar results... and would be fun to achieve what NVIDIA will no doubt try to claim is "Impossible" without their Revolutionary new Hardware.
Not to whom I was replying to and frankly the base topic of "I want to trade my AMD Card for an NVIDIA Card" on the AMD Support Forums is a pointless topic as there will simply not be any takers... even if they're Performance Cards unless he's willing to trade down, then he's simply not gonna have any takers.
Where-as a comparative against the "Next Gen" NVIDIA, is actually somewhat pertinent and likely more interesting to those on these forums.
"and would be fun to achieve what NVIDIA will no doubt try to claim is "Impossible" without their Revolutionary new Hardware."
I think it is already pretty widely known that it is possible, using more standard compute approaches. Interestingly, NVidia themselves had talked about Ray-tracing on Kepler using compute. Here is a slide from their SIGGRAPH 2012 slide deck.
So what happened? I think I have touched on it on this forum before, but most of what NVidia does with gaming hardware is through the lense of maintaining different product stacks between professional (scientific and modeling) hardware and gaming. They learned that lesson the hard way during the Fermi/Kepler days, when the original Titan Black was purchased as an entry level professional card effectively cannibalizing the much higher margin Quadro segment. After that, NVidia effectively scaled back the compute functionality in Maxwell/Maxwell 2/Pascal etc, as to maintain the distinction between professional and gaming cards, while maintaining they're bottom line in both market segments. Since the compute functionality was dropped, so was ray tracing and we never heard about it for six years.
As DX12, Vulkan appeared, NVidia developed Gameworks. DX12 and Vulkan both can utilize the higher compute functionality to generate spectacular graphics effects. Gameworks is designed to achieve those same effects using more standard gaming hardware. So rather than change their hardware and again make their gaming cards more attractive to professional buyers, they try to steer developers into using other approaches to enhance graphics which would allow NVidia to maintain a differentiated product stack.
With the launch of DXR, Microsoft is bringing ray tracing to the DX12 API. This can be done, by NVidia's own admission, via compute. The RTX cards then, are just more of the NVidia modus operandi, which is NVidia developing a specific thing, to do something that can be done with more traditional compute hardware. There is already and AIDA64 Ray-trace benchmark that utilizes an FP64 engine for ray-tracing, so why not go that route? Because an RTX core isn't going to work with professional software, so again the product stack is maintained. And if NVidia didn't put out and alternate solution to do ray-tracing, then developers might start to do it via compute (as NVidia suggested in 2012).
As you noticed, my blurb above is framed in such a way as to portray NVidia's interest as protecting their professional card profits. That may be the case, but you could also argue, effectively I might add, that creating separate product stacks makes the cards unattractive to professional users and keeps the gaming hardware in the hands of the intended audience
Regardless of specifically why NVidia chose the path they did, the outcome is that games tend not to run as well on AMD hardware. Since AMD is a smaller company, they can't afford to make a different die for each market space. Vega, ultimately, is a single die designed to serve all different market segments, which it does do surprisingly well. But with all the extra compute cores on those dies utilized for professional software, they generate far more heat in traditional gaming workloads. AMD based cards would actually fare better if game developers would utilize that compute hardware just hanging out on the die. But that is exactly what NVidia doesn't want, if developers start to use it, they would need to add it back in to stay competitive which it turn would once again blur the lines between professional and gaming market segments.
And so the fight has been for the developers. AMD has opted to put their hardware in consoles and NVidia uses Gameworks. AMD hoped that developers would develop a game for consoles first (and they do) and then port directly to PC. But usually, developers care so little about the PC port that they farm that out to a separate studio who operate on a compressed time table for release (see Mortal Kombat X, Arkham Knight PC releases). Those secondary studios then have to utilize tools like Gameworks to make the PC title launch on schedule, leaving AMD discrete graphics users with a massively deoptimized title. All that does is damage AMDs mindshare in the PC gaming community, as console users rarely actually know what is inside their machine.
But regardless of intent, the end result is undeniable. NVidia will actually leverage their market position to hold back the advancement of gaming. We could have had ray-tracing via compute six years ago, but NVidia's desire to maintain product differentiation hamstrung that effort. Only now that ray-tracing could advance down the compute avenue without NVidia do they throw their weight behind a different solution. And of course, the RTX solution only works on their new lineup of cards, whereas a compute based solution would work on their older hardware and AMD hardware as well.
with mining down the crapper, prices should be softer for used video cards by now
and i need specific make/model before i start peeling off a bunch of franklins
RE: This is a +15.5% Peak Performance Difference.
I do not understand where that % figure comes from.
Here is how I calculated the percentage difference:
((13.4 - 11.3) / 11.3 ) * 100/1
(2.1 / 11.3) * 100/1
0.185840708 * 100/1
18.584% performance increase of GTX2080Ti versus GTX1080Ti according to the figures reported by Anandtech.
I know how to calculate FLOPS.
I based my "guestimate" of gaming performance improvement on increase memory bandwidth and factored in how much a memory overclock helps benchmark performance on previous Nvidia cards I have looked at. I did not talk about overclocking anything.
My guestimate may be way off.
I think GTX2080Ti will be capable of 16.4 % Graphics Benchmark performance improvement versus GTX1080Ti.
How that translates into actual game performance may be lower. It often is.
We will only find out when the benchmark numbers are out.
RE: Well, given that the RX VEGA 56 is effectively on-par with the GTX 1070Ti in *most* titles and the RX VEGA 64 is effectively on-par with a 1080 with Overclocking again in *most* titles
I consider *most* Titles are DX11.
I trust reviewers like Gamers Nexus, Hardware Unboxed etc etc. They look at DX12 and Vulkan Titiles which should be in AMD favor.
They seem to think that GTX1080 still beats Vega 64 and GTX1070Ti beats Vega 56.
I do not know where you get your pricing from, but Vega 56 is still up against GTX1080 here and Vega 64 had generally been up against GTX1080Ti.
Prices have tanked this week though, for a few AMD cards, plus AMD has a free game promotion. Nvidia are also doing game promotions.
Vega 56 AIB cards I am interested in, 2 slot high cards, are very difficult to get. They are on pre order even now.
I suggest you look at this review, and then tell me things are great.
I am in process of purchasing a Vega 56.
I saw the above.
I have asked the reviewer some questions.
As for Navi. When is it turning up? 2020? I thought it is a mid range RX580 replacement?
RE: I might see about putting my nose to the grindstone with Radeon Rays over the next month or so, see if I can't coax my RX 480 to achieve some comparable results to what NVIDIA RTX is showcasing because frankly I actually believe it's more than possible to achieve some similar results... and would be fun to achieve what NVIDIA will no doubt try to claim is "Impossible" without their Revolutionary new Hardware.
Go for it.
Please create a separate post, and report the results.
It will be very interesting to see how you will be able to run Shadow of the Tomb Raider with Radeon Rays.