Usually the XY00 AMD and X0Y0 nVidia card are approximately on par, up to about the x080 level on the nVidia side. AMD's high end usually doesn't reach the ridiculous high of whatever the x090 can do, but AMD's top end is also far cheaper. AMD sometimes adds some extra SKUs to fill in between the mainline products, like the RX6650.
Cheaping out on VRAM is really weak sauce because it limits the lifespan and utility of the card. 8GB cards have been relatively mainstream probably for three full generations of cards at this point. They can't be the norm going forward.
Architecturally, everything that isn't an x86 PC will typically have a single pool of fast RAM that's shared between CPU and GPU and those two things will be on the same SoC. PCs can share RAM with a GPU as well but everything I've ever read about them doing that makes it sounds like it's profoundly inefficient. I understand that this is why PS5 game ports are uniquely demanding for Windows PCs.