| Intrinsic said: Then you are choosing to ignore the facts. XB1x with a 16nm APU has 6TF. Going from 28nm to 16nm allowed MS to increase their GPU CU count from 14 to 44. And sony to do the same from 20 to 40. But yet you feel going from 16nm to 7nm will only yield a 50% bump at best? That makes no sense. |
And you are ignoring some other facts. - It's not as black and white as you portray.
Remember... That Microsoft ditched the ESRAM with the Xbox One X... On the Original Xbox One, the SoC was roughly the same size as the Playstation 4, but had 50% less CU's... All because of that SRAM.
The move from 16nm to 7nm though will be a big one as it's a massive shift in geometries, 16nm is based on 20nm remember.
Where the extra gains will come from will be successive refined nodes... I.E. 20nm became 16nm with Finfet, which became 12nm and so on. Plus the refinements of each advertised process... Like 16nm Low-Power. etc'.
Same thing will occur with 7nm, expect it to be rebadged/refined as 6nm and so on.
TSMC pegs the move from 16nm to 7nm as 70% decrease in chip size with a 60% decrease in power consumption.
Global Foundries pegs the move from 14nm to 7nm as 65% decrease in chip size with a 60% decrease in power consumption.
That will of course improve as the process matures.
| Intrinsic said: Barring any other improvements, the PS5 and XB4 should at the very least have 12TF GPUs from just doubling the CU count from what we have in the XB1X which is something that is a default benefit of using 7nm fabrication. And this is assuming that clock speeds aren't improved in anyway. Like this is lookin at the least amount of specialized hardware engineering. |
The Xbox One X is a "Premium" console not designed to hit low price points like the Xbox One or Xbox One S though.
With that said... Regardless of this forums userbases addiction to "Flops". - A pair or 6~ Teraflop Radeon RX 580's can outbench a 20~ Teraflop Vega 64 chip anyway, meaning there is nothing wrong with 10TF base machines if they are an efficient design. (None of AMD's GPU's currently can be regarded as efficient though.)
I doubt we will get anything like those chips anyway. - The current whispers-along-the-grape-vine is that Navi will be a Polaris Replacement, essentially should bring Vega 64 performance to 150w TDP, perfect for the mid-range, perfect for consoles.
Expect 64 CU's (Or there-abouts), expect more conservative clocks to keep power consumption in check.
| Intrinsic said: Unfortunately this is impossible. Its not just a size issue but the nature of game data. A game may need to load say 4GB of data data into RAM. Some may be mistaken to think "hey we have NVME SSDs now capable of 2GB/s transfer speeds so let them just throw one of those in there", but even if that were possible it won't make much of a difference. Reason being that that 4GB is made up of thousands of really small chunks of assets. Access times end up being the limiting factor. Eg. a game takes 40sec to load with a 150MB/s HDD. 20sec with a 550MB/s SSD 17sec with a 2000MB/s NVME SSD. Best way to combat load times in games is via really smart data streaming... but not all games can work like that. |
There are ways to circumvent that... I.E. Smart data streaming like you alluded to.
For instance, during 7th gen, some games would send a burst packet of data into DRAM and then used a few CPU cycles to decompress and "unpack" that packet into the dozens of smaller files, that reduces disk access significantly.
Zones said:
You mean like the PlayStation VR? Oh wait sorry, that was clearly a blatant Virtual Boy rip-off. My bad. |
This is going to go nowhere, so best to end that discussion now.
--::{PC Gaming Master Race}::--
