The 7850 and 7870 were available in march 2012. The GPU architecture and functionality will be better than a 2016 GPU but in raw power it will probably be comparable to middle range 2016 GPU If the new standard for process node shrinking is 4 years and a half to 5 years...
Next process node will arrive in 2020 2021, for a launch at 14nm or 16nm in 2019 or 2020 it will be impossible to build an economically viable console more than 2 to 3 times more powerful than a PS4 if the slowdown in process node shrinking continue...
Something between a mid to high range GPU of 2016 just talking about Raw power...
the 28nm planar node was (is) long lived for a number of technical reasons. However, the next couple of steps look set to be rather quick. Samsung (GF) 14nm is already in production, as is TSMC16nmFF, and 16nmFF+ is ramping for full volume production as I write this. Now, these two processes are unusual in how much they share with the previous (and not very widely utilised) 20nm planar nodes. If you want to be uncharitable, you could say that TSMC and Samsung has taken 3-3.5 years to move to 20nm with FinFET transistors, but there seems to have been some additional improvements between 20nm planar and 14nm LPP, and 16nm FF+ respectively beyond just the transistor structure. Furthermore, both foundries look very committed to bring out the next full node lithographic step at 10nmFF at the end of 2016, which actually seems doable, although the number of process steps (and thus cost/wafer) will increase quite a bit. It may well take a while before 10nm is attractive cost wise to GPU manufacturers.
The next step after 10nm, 7nm, is trickier. Both TSMC and Intel seem to build production capability based on EUV. Extreme Ultraviolet (EUV) Lithography uses light of a shorter wavelength (13.5 nanometers) than the current standard in volume production of the most advanced chips, immersion lithography (193 nanometers). EUV can thus image smaller features without the need for multiple exposures, and allows semiconductor device makers to simplify the manufacturing process, exposing a critical layer of a chip in a single step. This can actually decrease cost/wafer, but it requires that the output of the light sources can be increased, to keep exposure times short and thus maintaining production rate in the lines. TSMC and Intel obviously thinks this will be achieved in the necessary time span, but it is not there yet. Also, FinFET may have played out its role, and new transistor designs will probably take its place which may or may not be a smooth process.
So at 7nm, the crystal balls get a bit murky. 10nm simply requires "more of the same" (hah!), while 7nm holds potential for both completely new cans of worms but also for smoother production with EUV.
Also, it bears mentioning that node designations is strongly related to marketing. A node change these days may not bring the overall benefits of a node change 20 years ago. So when for instance TSMC says they will start 7nm production in 2017, grains of salt should be distributed liberally all around when it comes to assuming what that actually means from a purely technical point of view, as well as the date.
