Please bear in mind you’re comparing a DX11 implementation with a DX12 implementation, both of which are really but really not managing VRAM nor assets the same way at all.
As it has been discussed in the past, the more VRAM your video card, the more RAM the game will use ONLY because DX11 maintains a RAM copy of every single buffer and textures you’re hosting in VRAM (well not all but it is just to keep it simple enough here).
It is also the reason why many (maybe not all though) have reported having better fps with 64GB+3090 vs 32GB+3090 because of the 3090 24GB VRAM…
[HOW-TO] Graphics Settings and Performance Guide - Discussion Thread - #62 by CptLucky8
PS: last but not least, the DX12 model is meant for the app managing the backing store, the DX11 is all handling this for you. This also come with additional benefits, such as DX12 being more capable to support streaming assets than with DX11 (hence as an extension of this capability the added DX12U Direct Storage feature).
PPS: in my opinion all the discussion about SU5 limited on PC because of Xbox is moot until they deliver the DX12 version on PC. However it is not moot in the sense they’ve probably tuned the engine for Xbox limits, which is in any way positively impacting the PC experience by the mere fact they’ve had to further optimize their structures and algorithms to make it fit.
However, tuning for the Xbox hardware is not good for the PC for a simple reason at least: 100% of Xbox is using AMD GPU, 85% of simmers are using NVidia GPU (at least in the Navigraph survey). And this is a factor which you can’t ignore if the most demanding part of the code is tuned specifically for the AMD GPU. It doesn’t sound like something evident at first, but here is an article showing how this matters in practice. For example if the game is optmized with the following in mind because of the Xbox, this might not translate to any particular gain for Nvidia/PC…
https://gpuopen.com/wp-content/uploads/2018/05/gdc_2018_sponsored_engine_optimization_hot_lap.pptx
etc… all these papers are about optimizations based on the actual AMD hardware specific architecture.