Come on people, be realistic, those specs are nothing but a pipe dream and definitely fake.
Since when has any of the console makers ever put together these sort of slides this far in advance of release.
11.2 TFLops (22.4 at half precision) with 3640 stream processors that's very precise info, for a chip that hasn't even been finished yet.
Also bear in mind that the 7nm MI60 which AMD has just shown us does 14.7 TFlops but that is at 1800Mhz and 300W TDP, I know some of you love underclocking but come on, be realistic to get that down to manageable levels to be able to go inside a console and not need a massive power brick to go with it. Remember that the PS4 Pro has an internal PSU.
Don't get me wrong I would love if these specs were real, but in my opinion there isn't chance of that.
Based on what do you believe the chip isn't finished, it's rumoured to be launching before the end of 2019, as such it would have to be LONG since finished by now in terms of architecture, general performance. It may well not have finished taping out by this point but it would be through the process with all design choices made and finalised well over a year ago.
As for Vega, Vega was a huge heavy compute architecture. ITt has a fairly large die slice dedicated to the HBCC, it has a large memory controller for 4 stacks of HBM, it has half speed 64bit support, it has int4/8 support, all of which take up a large amount of die space.
With Vega capable of all types of compute and a relatively small die they are free to completely ignore any kind of professional compute markets for Navi and make it a purely gaming optimised architecture. Frankly it should be easy to shrink a Vega giving it only FP32/16 support, no HBCC, no 4x HBM stack memory controller, no 200Gb/s infinity links, into something probably closer to 250mm^2. Now as you may have noticed from available information, Vega 20 went with 25% higher clocks at the 'same' power, or they could have used the same clocks at half power. So what do you get with a 4096, optimised only for gaming chip that runs at 25% lower clocks than Vega 20, something with oh wait, 11TF of performance, at maybe 100-120W. Keep in mind that having shaders capable of FP64/int 4/int8 will take more power than something capable of only FP32/16 so it would already be less than 300W by some margin at the same clocks with all those transistors removed.
So it could be a what ~60mm^2 8 core Zen 2 chiplet, also downclocked for maximum efficiency, could be looking at maybe 15-20W for the chiplet, 120W for the gpu, 30-40W I/O and you have not far off the same power as a PS4 or more crucially Xbox One X.
This is the biggest problem I have with your concept of what is possible. Right now today you can get an Xbox One X with 6TF of performance using about 180W full load using 16nm chips. 6TF on 16nm... but 11.2TF on 7nm, which is one of the biggest jumps in nodes ever, is impossible?
More over the PS4 had an internal brick and used 170W or so full load, what on earth has a PS4 pro using an internal brick got to do with anything. PS4 Pro using an internal power brick doesn't mean a PS5 has to at all, but more over, if they were happy with the absurd level of noise the PS4 made and the not quite as bad but still high noise the PS4 pro makes, why would that preclude them doing it for PS5, it's a completely illogical argument.
Fact is a fairly streamlined part Polaris part Vega Scorpio which also removes most of the 'waste' compute features within Vega 10 manages 6TF on 16nm, 11.2TF from Navi on 7nm looks not only possible, but pretty easy and actually somewhat unambitious.
THe thing here is the PS4 pro was PATHETIC, in power, it runs miles below what a console on 16nm was capable of which is exactly why the Xbox One X manages to get double it's performance within a fully working semi well designed case. Doubling GPU performance on 7nm is not a major task at all. Double the Xbox One X performance on 7nm should be the bare minimum people are hoping for in reality.
