The way I see it, one of the most prominent reasons for both Microsoft and Sony going with a very traditional arrangement of a PC-like architecture was to get PC-like perks, like making BC more accessible in future iterations.
These half-gens should keep the latter years' software sales flowing better than the PS360, so BC on the PS5 and XBTwo should be more important than the last iteration.
yea, I've heard it many times, went cheap with ddr3 & esram.
I think the decision for going with DDR3 + ESRAM was based on how much Microsoft wanted to make sure they had 8GB of total system RAM. Sony took the risk of only ending up with 4GB when they chose to use GDDR5, Microsoft did not.
I think the long-term costs associated with having to use 16 separated chips of some of the fastest DDR3 (on the way to becoming EOL'd for large-scale production BTW) won't end up being cheaper on the long run, and Microsoft's hardware team is fully aware of that.
Just like they were fully aware that bringing a substantially less powerful console to the gunfight would lose them some customers (either it's a substantial amount of customers or not is a whole other story).
That said, I'm sure there will come a day when the PS4 Pro is set free but in general it looks to be held back somewhat.
The leaked "Neo documents" and statements from Sony reps and developers don't seem to indicate towards any restrictions. I don't doubt there will be some learning curve to take full advantage of the Pro's architectural improvements (e.g. 2*FP16 processing), but I don't know of any limitations being imposed to the Pro, at the moment.
I also don't believe MS would wait a year and put in something older than Polaris / Vega if Sony already has a Polaris hybrid in their system
From Cerny's statements, I think you mean a Polaris-Vega hybrid?
GDDR5X at a very modest (for GDDR5X) 10 gHz would deliver 320 GB/s on a 256 bit bus, and indeed this was nvidias preference for the 1080 so 384 bit certainly isn't MS's only option ... though it may well be the cheapest and easiest to engineer with AMD.
I think the 12GB on a 384bit bus idea came from the fact that the rendered PCB in the reveal video showed 12 RAM chips, and not from the bandwidth numbers.
Once you go over 8GB of GDDR5 - which Scorpio really needs to anyway - moving to a 384-bit bus is likely the cheapest option.
They could use an approach similar to the Pro and implement some slower memory through the southbridge for app switching + certain system tasks while simply unlocking more of the GDDR5 for the developers. Truth be told, for game developers it matters little what memory is being used for system tasks.
Why are people are assuming they wont offset the initial launch price of Scorpio to win huge market shares? They would make it up through the sale of a couple games and increasing their consumer footprint.
By late 2017 the current gen will probably see some >70M PS4 and >40M XBone sales, not to mention the Nintendo Switch possibly taking away some more of the "console budget" from the populace.
I wonder if the
winning huge market share expression will even be applicable by then.