Makes sense, bigger die = more transistors crammed in. I suspect they will stick to around 500mm² as maximum, as they have never made a larger die than that.
And RDNA 3 will probably follow similar rules, but with multiple GPUs per graphics card.
Regarding the TDP of desktop GPUs, I wonder how much TDP the ray tracing hardware will require for RDNA 2? If it's like 50w, that could be pretty crippling for performance of desktop GPUs. I'd guess 20w..
Apparently, the max TDP is around 375w for a graphics card with 2 8 pin connectors, so maybe a RDNA 2 GPU with over 300w is possible.
Dual GPU HD 6990 here:
https://www.techpowerup.com/gpu-specs/radeon-hd-6990.c275
Kinda makes me cast doubt on the need for a new PSU connector type for Nvidia GPUs, especially since a lot of half decent power supplies (like Seasonic Focus series) have 4 x 6+2-pin PCIe connectors.
I think the only reason not to go over 300w TDP is GPU temperature. The 5700 XT typically hits
~76 degrees celsius under load @ 225 watts, which isn't exactly cool.
If total power corresponds to temperature, a similar RDNA v1 GPU with a TDP of 300w could hit
101 degrees (76 + 33.3%). Apparently the 5700 XT throttles it's performance if it reaches
110 degrees, e.g. pretty damn hot.
So, I think an RDNA 2 GPU with a TDP of 300w would have similarly high temperatures, perhaps between 90-100 degrees. If increased to 320w TDP, it could reach
108 degrees (76 + 42.2%). I'd say that's too high and could reduce the GPUs lifespan and increase product return rates potentially.
Who knows though, perhaps they've made some improvements to heat dissipation / cooling with RDNA 2 GPUs.