In testing that basically amounts to dick, it's for a lack of better words marketing fluff to offset teraflops to a higher figure.
Getting developers to adopt AMD's instructions that are compatible with dual-issue should happen much more often in the walled garden that is console development.
It's obviously never going to behave like a 2x RDNA2 GPU because there's no doubling of registers, caches, bandwidth, schedulers, etc. but it still might amount to considerably higher average performance.
If it's "just" 25% higher performance it's still a massive win considering how little amount of transistors AMD spent on dual-issue FP32 compared to RDNA2.
Which again makes it marketing jazz, the reality is still the lesser figure in practice.
Remember Rapid Packed Math from AMD? Yeah, same level of nonsense.
Rapid Packet Math was far from nonsense and pretty much allowed for free linear upscaling like FSR1.
That's why the old Polaris cards would barely get any performance uplift from using it, for example.
Whats the jump in power and efficiency from rdna 2 to 3? Like 30%
Unfortunately it's almost none, but mostly because AMD's RDNA3 implementation on TSMC N5 was bugged and the power/clock curves were similar if not worse than RDNA2 on N7.
As weird as it sounds, the Navi 33 made on the older N6 gets much higher clocks than Navi 31 and 32 on N5.
Again, base PS5 is how they will hit their sales targets, it's on sale right now infact. The Pro will be positioned as a premium product with a decent margin built in like PSVR/PSVR2, Dualsense Edge, and PlayStation Portal.
Again, PS5 Pro still needs to sell
some units to justify the investment into a new hardware platform and the software dev efforts to implement its optimizations (heavier raytracing, PSSR, etc.).
If the PS5 Pro sells less than e.g. 500k units worldwide, then it's absolutely a disaster for Sony, regardless of how much money they're making per console (which Sony never did with any new console launch, so that was a failed prediction from the start anyways).
