Sony is focusing on Hard-Core Gamers for PS5

[Trumpstyle]

drkohler said:

Pemalite said:

Once you hit 1GB of used capacity on each chip, then you can only start filling up the 2GB chips...

And this is when you switch from clamshell mode to non-clamshell mode on the 2GByte chips on both channels, putting the 1GByte chips onto high impedance.

As you say, it is really very complicated implementing such shenanigans and I doubt they are going to implement this. 

I see either 16Gbyte on a 256bit bus or 24Gbyte on a 384bit bus, leaning towards the 16GByte solution which is obviously much cheaper. Also with the "superfast ssd" solution combined with improved memory compression technology leads me to think 16GByte will be enough memory for a console.

Pemalite said:

drkohler said:

And this is when you switch from clamshell mode to non-clamshell mode on the 2GByte chips on both channels, putting the 1GByte chips onto high impedance.

As you say, it is really very complicated implementing such shenanigans and I doubt they are going to implement this. 

I see either 16Gbyte on a 256bit bus or 24Gbyte on a 384bit bus, leaning towards the 16GByte solution which is obviously much cheaper. Also with the "superfast ssd" solution combined with improved memory compression technology leads me to think 16GByte will be enough memory for a console.

Microsoft's E3 trailer shows us that Scarlett is possibly using a mix of 1GB and 2GB chips.
Digital Foundry mentions it here even: https://www.eurogamer.net/articles/digitalfoundry-2019-project-scarlett-spec-analysis

Whether they employ a 256bit or 384bit bus remains to be seen... And again, I am not asserting anything as fact, just hypothesizing on possibilities.

As for the SSD... That isn't a replacement for Ram.

With the recent "Flute" leak for PS5 I'm pretty sure I know what Microsoft is doing mixing 1 and 2GB Vram. As you 2 said, you can't mix those memory types and run them at full speed. So the only solution that make the most sense is 18GB of VRAM on a 256-bit bus, basically you use 8 2GB Gddr6 chips which give you 16GB of vram and run them at full speed, than you just add 2 1GB ggdr6 chips to those memory controllers and run them at slow speed and use this for the OS. This give 16GB Vram for games at 448 GB/s memory bandwidth speed and 112 GB/s with 2GB Vram for the OS. With a SSD they probably won't need more than 2GB for the OS.

Second solution I come up with which is unlikely is a 320-bit bus with 12GB Vram, this gives 560 GB/s memory speed, you use 8 1GB vram chips + 2 2GB Vram. You can use something called FLEX MODE and use all 10 Gddr6 chips and run 10GB Vram at full speed and rest 2GB Vram at slow speed. Which gives 10GB available for games and 2GB for the OS. Only reason I see Microsoft doing this is either Gddr6 chips are very expensive or they doing a powerful gpu with low amount of memory but as I said the first solution is very likely what microsoft is doing.