I thought we laid it to rest that the unified memory really makes the dedicatee ram question largely irrelevant.
Totally off topic but well... not really.
There's some interesting tradeoffs that I still haven't seen good data on. Gonna break this out into three parts:
Dedicated Memory, Shared Memory, Unified Memory
Dedicated Memory and Shared Memory function largely the same except that in shared memory instead of it's own dedicated hardware it sections off a chunk of RAM from the main memory hardware. Each has to copy things from "main" memory to video memory. The difference is when looking at dedicated GDDR7 you have a 1.5TB/S data pipeline to copy where as with shared LPDDR5 you're looking at about 500GB/s (and it's the same memory bus which slows things down further).
Unified memory is different than shared because instead of sectioning off memory it is all stored as one which means the CPU cores and GPU cores can access the same addresses. This is cool because you don't have to copy stuff between main and video memory. No copies speeds things up but it's still 500GB/s.
So the question is "which is faster?" If you have a workload that offloads a bunch of shaders and then proceeds to do most processing on the GPU the dedicated is going to be way faster because there isn't a lot of copies. If you are regularly switching out what would go into video memory the unified memory might very well be faster. And I haven't seen any good testing on this front to say one way or another.