Currently Online 
Lafiel said:
As I said, my theory is that it's simply not viable to shrink the current architecture, so they have to use one made for 14/16nm finfet to use this process. But using a same-spec (18CU etc) chip as the old one opens a whole new can of worms, because code written "to the metal" for the old chip won't run as efficiently out of the box (and they certainly don't want every game to need a patch) on the new architecture, so to circumvent this problem they imo always had to throw in additional CUs to make sure the finfet chip can brute force PS4 parity in unpatched games. Doubling the CU count even makes it possible to sell the PS4 Neo at a higher price as long as both version coexist > higher margins. Due to the enormous differences in structure size between chips made in 28nm and 14/16nm the PS4 Neo APU will certainly have a smaller die size than the current PS4 APU and in the end (when yields have caught up and process costs have sunken due to enough production capacities being available) that's the most important factor for chip price per unit (the more working chips you get from a silicon wafer, the cheaper they are per unit). |
Ahh that makes sense, so the new GPU would be a different architecture. Sounds like a massive oversight at the time the ps4 was designed, causing quite a pita for developers going forward. So brute force emulation of the base model with the option to write more efficient NEO specific code. (those separate NEO GPU binaries make sense now) Clusterfuck with a silver lining...
