Why would you expect a N4 chip to have perf/xtor more in line with N6 than N5?
We don't really have an N5 chip from AMD with which to make this comparison because N31/N32 are MCM. The closest is Zen 3 to Zen 4 which saw a 58% transistor count increase deliver 38% more performance in Spec Int nT but some of that budget was spent on AVX 512 where the performance increase is infinite since Zen 3 does not support it so its not quite as cut and dry but lets assume Spec Int is similar to raster and the additional transistors get spent on RT performance improvements just to give us a rough starting point.
That would mean scaling from N21 to a monolithic N5 class design 42.3B transistors should provide about 38% more performance which is inline with the 7900XTX if scaling up from the 6950XT. Issue is the 7900XTX uses 58B transistors. Even if you take off the approx 2.4B transistors used for the MCD GCD phys that would not be needed in a monolithic design it means you would still be at 55.6B transistors which is more than double for just 38% more performance. It is utterly terrible. If we take the 4K delta between the 7900XTX and the 7900XT then it would mean with linear transistor scaling 7900XT performance should be doable in 33.8B transistors if AMD can match the performance to transistor increase ratio of Zen 3 to Zen 4. In 240mm that would need 141M xtors /mm.
I don't even know if that is a valid way to look at is given the differences between CPUs and GPUs but without a monolithic 5nm GPU from AMD as a starting point is the best we have.
I could look at NVs change from Samsungs 8N to their modified 4N. There it took 26% more transistors in AD104 to match GA102 at 1440p (although it does fall behind at 4K). If we start with the 6950XT transistor count and AMD do similar to what NV achieved then to match 6950XT performance it would 33.8B transistors. If we start with N33 then it would take 2x 16.8B transistors to match 2x N33 performance which is 6950XT / 7900GRE level for 33.6B transistors.
All very similar transistor counts for between 7900GRE and 7900XTX performance when scaling based on what has been achieved by other vendors or with other chips. Not sure if it is indicative of anything or if it is coincidental though.
I think what it points to is that for N48 to deliver this kind of performance it does not take a transistor density that AMD have not achieved or a scaling factor that NV / AMD have not managed elsewhere so there is nothing that has not already been done.
If we go back to AMDs prior die shrink it was Vega 10 to Vega 20 where a 5% transistor bump led to a 20% performance improvement. If AMD managed that N44 would be around 14B transistors and perform near the 3070Ti and N48 would be 26B transistors and perform like the 7900XT. The density required for that with the supposed die sizes is just 107M xtors / mm though so AMD look to have density headroom if N4P requires more transistors to hit that kind of performance.
The final way I am going to try and have a look at it here is via N32. A monolithic design here would be in the region of 34.5B transistors if you remove the MCD GCD PHYs. This would require a density of 143.75M xtors / mm to achieve so is on the high end of that spectrum but the real question is do you think if plausible that AMD won't improve perf/transistor vs what N32 achieved? My personal view is if they fail to achieve that Radeon group might as well give up on dGPUs and stick with the CDNA lineup.
All the ways I try and come at this to get to a performance ballpark land me in the 7900GRE to 7900XT performance window at transistor densities that range from just about seems doable to seems pretty easy to do. Now clearly AMD may fail to execute again or Radeon group can fail to achieve what NV / Ryzen group managed to achieve when moving to N5 nodes but that just feels like assigning a level of incompetence to them that I don't think they deserve as yet. They can prove me wrong of course.
