maybe oberon is 300mm2 at 7nm+. that would mean it was 360mm2 on 7nm. should line up with the bigger navi rumors.
300 mm2 means a 40 CU gpu at 251mm2, a 40 mm2 cpu and 9mm2 for everything else. maybe 9mm2 is possible on 7nm. the hd 7870 was 212mm2 and the jaguar was 70mm2 so everything else in the ps4 apu was 68mm2. a 4x reduction in node size and we are looking at 17mm2 for that everything else. maybe hbm controllers take up less space on the apu? i cant seem to get an answer for this. i do recall hearing that the vega 7 gpu was so big because of the wide hbm2 bus.
regardless, that 300mm2 calculation does not take into account any rt hardware that might take up space on the gpu. rough estimates here pegged RT hardware at 10%. so thats 25mm2. how do you fit that in the 300mm2 oberon budget?
if the 300mm2 rumor is true, it has to be on 7nm+.
No.
HBM2 and HBM2E can be stacked around the chip. In GDDR6 if you look at the PCB, memory is all over the place.
Radeon VII with HBM2, you will see gpu surrounded by HBM2 stackes on same dye.
Here's a 2080ti with GDDR6 notice you see the chip, then see memory banks farther around it not on the actual dye.
And for the record I don't think final hardware is using HBM2. I believe their devkits them have are using HBM2, but final will be using HBM2E. I believe the recent info with HBM2 chips being really cheap was because HBM2E was starting to go into mass production on top of there being demand for Radeon VII, and couple render cards with Datacenters recently being built for AI farms. WHich use HBM2 memory.
