Oh I am aware its not as clear cut as that. But I also know that the move t 7nm is more significant than the move to 16nm from 28nm. I aso know that with 7nm building an APU like I have described is possible. Especially if they are aiming t have a similarly sized APU in those consoles as we do now in the PS4pro/XB1X
The move from 28nm to 16nm was actually extremely substantial, in some ways... More substantial than the move from 16nm to 7nm... And that is mostly thanks to Finfet. Aka. "3D Transistors".
50% higher clockrates, 50% smaller chips in conjunction with 50% less power consumption wasn't uncommon when compared just against 20nm. (Let alone 28nm.)
And that is before we look at variants like 16nm FF Compact.
The issue with 16nm though is that it is based upon 20nm BEOL... So they were more reliant on Finfets for gains. - But when you start comparing the process against 28nm, it's a significantly more substantial jump.
I strongly dont think the things that make the XB1X "premium" will be issues when looking at next gen hardware. First off, I dn't think the XB1X APU has anything to do with it costing more. It has a GPU with 4 more CUs than the PS4pro but the overall die size is very similar to the PS4pro APU. It using a Vapor chamber and 12GB RAM on a 384bit bus is why it costs more than a PS4pro.
I agree, to an extent.
When you start to break down the Xbox One X into components and analyze what makes it a "Premium" device, you can quickly see where they can cut some corners or substitute some technologies in order to allow for lower price points.
I dont see them using a vapor chamber r even a 384 bit bus n next gen consoles.
Yeah. I think for a base console, Vapor Chambers will be relegated to the scrap heap. - They are simply to expensive.
That is provided they stay around the 120w TDP mark... Which 7nm should allow for. - Navi with 64CU's should fit that bill, in-fact the rumors are suggesting that is exactly where it should fall.
A higher end console like the Xbox One X though that pushes 250w~ TDP though will need a vapor chamber.
They just don't need to and would still be able to get performance from their components that dwarfs what we have with the Pro or the X. Eg using 12 GDDR6 chips on a 256 bit bus will still yield over 600GB/s of bandwidth or if using a 384bit bus will result in around 800GB/s. And assuming they are using the cheaper 14Gb/s/pin variant of the memory.
Dwarfing the Playstation 4 Pro or Xbox One X was possible before those consoles even released, it's not actually a difficult task... All just comes down to costs in the end.
As for GDDR6, that is likely a given, Microsoft probably won't make the same mistake twice and go with an older/slower memory technology that is getting more expensive... And Sony will likely not go for something more exotic/expensive again either.
256bit bus is the sweet spot for a base console, it's a known quantity, doesn't require an exotic memory controller, doesn't require a ton of PCB traces and thus layers and doesn't require a complex power delivery system to feed everything.
512GB/s of bandwidth on a 256-bit bus is fine, great for 4k gaming and a base console... And perhaps we might see an iterative update to Delta Colour Compression.
Tonga's colour compression bolstered bandwidth by 40%. - Polaris increased that by an additional 17%... So we could be looking at 839GB/s of usable bandwidth in an ideal scenario... And that is before we start taking into account modern improvements to that technology that could add another 10-20%.
And the fact that Navi should also bring with it functioning Draw Stream Rasterization and Primitive Shaders to further increase efficiency. (With some luck.)
While I am aware f some sort of supposed 64 CU limitation with regards to Polaris we don't really know if that limitation will carry through to navi so I don't think we can just put the ceiling at 64 CUs right now.
64CU's tends to be the hard limit for Graphics Core Next in general, not just Polaris.
Navi is supposed to have "improved scalability". - So AMD may have circumvented that, then again... They might not have and would rather put a greater focus on their next-gen GPU architecture rather than Navi which is the old dregs of Graphics Core Next.
64CU's is just going to be the mid-range, sweet spot at 7nm, it's where Navi is targeting... And the mid-range is where consoles typically sit on the GPU ladder.
Also the thing about conservative clocks can sound misleading. Even though you are right. Eg Say the CPU is based on an 8Core zen that ideally clocks up to 4GHz. Conservative clocks for consoles will probably bring that down to 3Ghz. Say the navi based GPU is designed to run at 1.8GHz - 2Ghz, console speeds will again be like say 1.2Ghz - 1.5hz. That is still a massive bump in performance when considering all the new components being used.
Conservative clocks is just so they can get as many functional chips out of a wafer. Not every chip will hit high clockspeeds with moderate voltages, so you have two options... Increase the voltage of every chip which increases heat, power consumption and the need for better cooling or decrease clockrates.
Graphics Core Next has proven to be a bit of a power hog at high clocks as well... For all the shit Vega gets, once you underclock and undervolt that GPU, it's actually really efficient, AMD just pushed things to far on that design.
For clockspeeds... Polaris was 150w TDP @ 1152Mhz.
Vega 64 was 1247mhz @ 295w TDP
I think around 1.5 would be a good ballpark to expect as 7nm brings with it some gains on that front. - I wouldn't even be surprised if it sits closer to 2ghz on the GPU clock.
Zen is interesting.... A 4 Core/8 thread Ryzen 2700u can sit at around 2.9Ghz on my laptop with all threads loaded... That is at 25w TDP.
If next gen goes with a single CCX and if AMD increased the CCX to 6 or 8 cores, I would expect 3ghz at around that same level.
That's what I mean by conservative... And not something like 4Ghz.
Yes, and maybe those methods can be employed again next gen but I think they didn't use those methods this gen because of the weaker CPUs in the current gen consoles.
This generations CPU's are significantly faster than last generations and offer better all-round performance, especially once we start looking at integer performance.
The leap wasn't as great as it could have been though... But that was AMD's own fault for not having a compelling CPU design to start with. - But that's no longer a problem thankfully.
1. Vega GPUs are already well above 10 TFlops and Navi will be the architecture used in PS5 which should be more power efficient. An 11 TFlop GPU is atleast expected of these consoles.
Flops is pretty irrelevant.