You fit it without a problem.
8 Zen 2 cores are 74mm² with full L3 cache. Having 1/4th of cache gets that chip to ~35mm², as more then half of die is reserved for cache.
RT cores in Nvidia cards take 8% of CU (not total GPU die), while Navi CUs use ~90mm² of total 251mm² die.
So even doubling RT per CU space vs Nvidia, you would only use additional ~20mm² over full Navi ~ 270mm².
Since PS5 would have 256 bit bus, and not 320 bit like Scarlet, you would save ~16mm² there.
Full size, Zen2 (1/4 of cache) + 40CU Navi (includes IOs) + RT solution that takes double space that of Nvidia on 256 bit bus die ~ 305mm². Still left with die space...
My thinking before Flute was, no way can they fit everything in there at 316mm. But with Flute having 1/4 of Zen2 cache, it all made sense. You could probably fit in 48CUs in 316mm² IF RT cores are similar percentage of CUs as in Nvidia case.
Compared to Scarlet, you would already dave 16mm² on smaller bus, but with faster memory you would hit similar BW. Perhaps Scarlet has 1/2 of L3 cache and then, its pretty much same amount of CUs and RT cores but with wider bus and slower memory, as well as more cache for CPU.
Would make for great chip in PS5 case IMO, although Scarlet would be having a bit more headroom IF really required.
BTW I dont know where you guys got 70mm² + for Scarlet. That would make it almost 400mm², and I think, based on die shot - its closer to 350mm².
Last edited:
goddamit, forget retail price! 2080 super probably cost around 200$ to make if not less, and even then you compare APU vs GPU, unified memory vs not unified memory, optimization vs brute force, one configuration vs many configurations that advantage for consoles, and that's why PS5 perform like 2080 super or even better with less actuall Tflops it always been this way always will.
