It is always necessary to compete at the high end. The vast majority of the public doesn't do a lot of research on which models perform best and in what titles. If NVidia has the fastest high end card, that leads to more downstream sales, regardless of how competitors behave due to the belief that "Nvidia is the fastest". AMD has been falling further and further behind in this category since GCN launched in 2011. While the 7970 and R9 290 were the fastest cards on the market when the launched, the Fury X only managed to equal NVidia's Titan and 980 Ti. Vega now, only managed to fall between the GTX 1080 and GTX 1080 Ti (closer to the GTX 1080).
That could actually be a boon for AMD. AMD cards tend to perform better with DX12 and Vulkan engines vs NVidia counterparts. While DX11 performance sees large gains for NVidia cards. If Intel's architecture performs well, and utilizes more of the compute orientated approach vs GameWorks, that could give developers incentive to develop games that utilize the Intel features, causing AMD cards to perform better in the process.
The new generation mid range performing as well as the previous high end would indeed be new. Polaris when it launched only performed as well as the R9 390, the high end two generations removed. It was still substantially slower than the Fiji based cards launched the previous year.
I haven't forgotten about the 1170 and the 1180, but in reality it may not matter. As NVidia kills production of the previous generation cards as soon as a new one launches, the 1070 and 1080 will be gone. If the 1170 and 1180 perform as well as the 1080 and 1080 Ti, NVidia may price them accordingly, leading to zero alteration in the GPU landscape. It would only be disruptive if NVidia priced the 1170 and 1180 exactly the same or lower as the current 1070 and 1080, and with NVidia that is a big if.
The new nvidia cards are going to be a bit different, including the memory control.. I feel and have always felt Vulcan is a crutch for amd gpu's. I seriously think Nvidia is not even remotely concerned with Vulcan... The gddr6 will improve performance across the board and also improve tdp across the board.. The NV cards also have a die shrink to 12nm which will also decrease tdp and increase performance.. The question is how much.. Rumors I'm hearing is the 1180 should be about 15-20% faster than the current 1080, with a lower tdp, which puts it close to the 1080ti in performance.. I want to see how the tensor cores are going to affect the high end gpus in the future.. The tensor cores are there for a reason, are they even being utilized yet? Pricing will be the biggest factor in these new generation gpu's as we know performance will be there..
I'm interested in both manufacturers gpus that are coming.. If AMD can put out a gpu thats runs right with the 1080 for $250 with a lower tdp than the 1080, that would be a boon to the mid range market for sure, but also a love fest for cryptominers, depending on what NV has on the table.. Amd wouldnt be able to keep them in stock, as they cant keep gpu's stocked now, not sure either company can keep up, or memory manufacturers can keep up, in turn that will dictate and increase prices due to lack of manufactured memory modules.
I bet the new nv cards will do dx12 just fine and likely excel at it.. My 1070, 1080 and 1080ti all ran and still run dx12 just fine.. My current msi 1080ti trio is amazing, runs like a raped ape and stays under 70c under full load..
I can also say I just bought an rx580, I previously owned a 480 and sold it last year for a huge profit when the mining boom hit.. The 580 does mine certain coins very well and thats the sole purpose I bought it..Although I do find setting up amd cards are more difficult than an nvidia rig build..
Like it or not, you can not take mining out of the equation, power to performance ratio is where its at with the miners and they'll buy many many more gpus than gamers ever will, look at the vega 56 for an example.... That will also become a problem as miners will start selling off their used gpu's, they'll flood the market with used 1070, 1070ti, 1080's.. It will also cause the price of new 1070, 1070ti, 1080's still in the marketplace to drop dramatically, which can be a good thing..
We all could be winners, lets hope so..
Vulcan isn't really a crutch, it is an open API that not only runs well on AMD and NVidia GPUs, it works in OS's outside of Windows. APIs like that would truly equalize gaming across platforms. Vulcan doesn't actually run worse on NVidia GPUs than DX11 or DX12, it just runs better than either of the other APIs on AMD hardware. As a fan of open source and competition on a level playing field, I can't really support NVidia. GameWorks was bad enough but the hits just keep on coming.
Essentially Geforce cards run better than AMD, they do so in large part because NVidia leverages it's market position to get developers to use rendering methods that favor their hardware, and then prevent AIB partners from selling competitor cards under premier brands, which improves their graphics performance and marketability and increases their market share and the cycle continues. I loathe that sort of behavior, as it leads to market stagnation and prevents innovation from anyone trying to break into the market.
I am really hoping that Intel, with their own massive market clout entering the fray and break things up a bit. I don't really believe that a flood of aftermarket NVidia GPUs will do much for prices. GPUs potentially used for mining are terrible investments and most gamers stay well clear of those. NVidia typically allows production of the previous generation to stop well in advance of a new launch, to allow the channels to clear.
I don't really follow NVidia news about their new architecture but a die shrink and different memory hardly sounds innovative. It would be nice to see HBM2 and some interesting ITX form factors, oh well (Maybe the Vega Nano will appear some day). Also don't expect any amazing FP64 performance, NVidia would never risk cannibalizing their workstation margins with a gaming card after the Titan Black fiasco in the Fermi/Kepler days.
Intel is having headaches galore at 10nm, 7nm is so bad even excedrin cannot handle it
even by 2020 may be too early as work with a wide range of equipment over and above steppers is needed
then there is the move to 5nm which is way out there is lala land