This is something pretty critical I dont see many people ever talk about(here or elsewhere). Ignoring the nonsense that MS were taking a $200 loss per console, there's still the undeniable fact that MS used a 360mm² die with obviously worse chips-per-wafer than PS5, and they never made any effort to port to 6nm to help with this, nor done any notable cost cutting revisions for the system as a whole. And their SSD setup was much slower, but instead of having that be an advantage in terms of cost, it's likely offset a fair bit by being 2230 form factor.
Really seemed like they were just taking the punches of not just having worse sales, but also having worse profitability, and doing nothing about it. I know Series X and S are both quite well designed consoles already without a ton of room for internal packaging improvements, but surely there's room to do something. Anything.
A few of us here have talked about way to possible cost reduce the Series consoles, particularly the X.
For example, moving to a 256-bit bus with 18 Gbps per pin memory. This would reduce the die area needed for 64-bit of PHYs and for two of the ten memory controllers (along with 1/5 of the L3), and physically remove two of the memory chips. This would allow a smidge more BW, free up die area and allow a smaller and possibly less complex / expensive mobo with fewer metal layers. (Part of the reason for a split board might have been to balance a more expensive board for the main chip with a cheaper one for the SSD etc - MS spoke of challenges around ensiuring signal integrity for the 320-bit bus).
On the down side, the faster GDDR6 would need more power per chip and be a little hotter, and the loss of 1MB (or 1/5th) of the
L3 L2 might not be compensated for by a few percent more main memory BW. Perhaps larger
L3 L2 blocks with the other controllers would be needed?
The current Series X die has GDDR6 PHYs along 3 edges of the chip, and these don't shrink nicely along with transistor density on smaller nodes. So maybe the current memory arrangement doesn't allow MS to take full advantage of a move to 6nm without a more extensive and expensive redsign of the chip than is deemed cost effective. Maybe that's been part of the holdup.
If MS could get down to a 256-bit bus on a 6 or even 5nm chip this might allow them to save on SoC cost, mobo cost, memory cost, cooler and psu cost, and the cost of the chassis (going to a single mobo instead of split).
Even a budget card like the RX 7600 (non XT) with a cheapo cooler uses 18 Gbpp memory now. I think it could really help MS across the board if they could move down to a 256-bit bus.
