DemonCleaner
Member
Actually, my die area estimates are based off Liara Brave's numbers from ResetEra. He takes RT area into account. Incidentally it isn't expected to be much:
a) Because the area devoted to RT cores for BVH intersection acceleration in RTX aren't much (Liara Brave increases each CU footprint by the same proportion)
and b) from AMD's patent on their own projected RT hardware implementation, it appears they're putting the BVH acceleration hardware in the RBEs, while at the same time it appears not to be as comprehensive in design as RTX's implementation. So the prevailing estimation is that it should take less area.
In which case, my die area estimates would be considered somewhat conservative.
In any case, that post was merely intended to be indicative of the relative difference in per die cost for going with a larger die with more shader cores, than a faster smaller die. It's only intended to be indicative.
do you have a link to the forementioned estimates?
i've done a layout proposal for a potential big next gen die last summer. back then not taking into account ray tracing components and came to very similar numbers for die size as you quoted. you seem technically interested, so i strongly suggest to read the first two posts (and maybe the last few).
i know the raytracing patent and came across some additional info on rdna design since then. i might update my layout prediction with this information when i find some time. anyways i don't think the surfaced facts still support the conclusion that we will see a <=370mm² die for XSX. It will be larger in the forementioned 400mm² zone. ps5's die accordingly.
Last edited: