I think this whole memory talk is being overly complicated. The way I see it is this
GDDR5 currently has a chip peak capacity of 1GB. So if you want 8GB/12GB in whatever application you are doing you need 8/12 chips on the board respectively. Its possible to have them arranged in a clamshell format (having a chip on either side of the board which literally means you can double the density of chips you can have in the same sized board but this adds complications and cost to overall design and efficiency isses to the chips performance. Not enough to break things but enough to ideally be avoided if they can.
GDDR6 is all of the abpve but with a higher chip capacity of 2GB/chip, more pins allowing for higher bandwith per chip and thats basically it.
So theoretically, if say the PS4 is beiing built with GDDR6, using the same 8 chip array we see in the PS4slim/PS4pro boards to get 8GB of GDDr5 we would instead have 16GB of GDDR6. The XB1X will have 24GB of GDDR6. Bandwiths will all be significantly higher too and i think its as simple as that.
And this is going with the most straightforward design implementation of GDDR as opposed to any complex clamshell nonsense, which in truth can double capacity.
I don't know why some would think its unlikely then to have as much as 24GB of GDDR6 in the PS5/XB2.







