Cross posting this from the DF thread on the xbox hardware.
Assuming it's an accurate image, the 7nm chip we saw back at E3 is roughly the same physical size as the 16nm chip seen in the X - likely a touch larger. Transistor density is unlikely to have doubled, meaning that the primary route forward for increased performance is frequency - and lots of it.
i have a pretty big issue with this quote in the article. why is it unlikely?
the 36 cu polaris was 5.7 billion transistors at 14nm and 232mm2. the 5700xt is 10.4 billion transistors at 251 mm2. it looks like it doubled to me.
The scorpio apu was said to be 7 billion transistors at 16nm. that includes the jaguar cores. we know the zen 2 is 2 billion, but almost half of that is taken up by the cache which the console makers would cut down as shown by the flute leak. lets assume MS only cuts down the cache by half. we have a 1.5 billion transistor cpu.
we dont know the CU count of the scarlett, but even if we go by oberon 2.0 ghz clocks, we are looking at a 48 CU GPU with 4-6 disabled CUs. so 54 minimum. how is that not over or around 14 billion transistors? or 2x scorpio?
i saw phil complain about moore's law slowing down as well. and yet in just two years since scorpio's launch, he was able to release a 12 tflops gpu based on a completely different architecture while doubling the transistors which i believe is what moore's law is all about.
whats even more crazy is that DF themselves point out how the chip is as big as the scorpio which was 358 mm2. the 40 cu 5700xt is 251mm2. the zen 2 is 70mm2 and should be around 50 with half cache. that leaves 60mm2 for memory controllers and a whole shit load of CUs. yes, the clocks will be higher but you always get clockspeed increases when you go down nodes. thats on top of transistor density.