Ok ... Can we fix either the 3d displace or make the 2d displacement support UDIM ?

[Bigguns]

3d displacement just can't be used for character since when it's far form the camera it has less details so it mess up my look dev .. plus it take humongous amount of ram and freeze my system.. trying to get half the quality of the 2d.

Right now, I'm using subduivion finally got it working with my udim set up and at least it's constant from far or close. But still, takes more ram then 2 displacement and is not as crisp.

And no, Zbrush don't use CK has subdivision method, they have their own thing.

What do you mean by deofrmation-invariant ?

[^Lele^]

2D displacement is sensitive to mesh curvature and (used to) have a parameter called "precision" to deal with more or less curved areas.
Regardless of how precision works now, 2d displacement creates an abstract heightfield representation of the base mesh, the resolution of which is decided in the settings, and of the original geo nothing is left at rendertime (and only at rendertime!).
So deforming geo will invariably suffer from popping and errors, as the evaluation of said surface to a heightfield will change based on mesh topology, per frame.

3D Displacement, and all the more Subdivision with a displacement map, particularly when set to world-space, purely take the base gemotery and create more triangles within the bounds of existing ones, and can be set up to prepare said mesh statically, and across all cores before the rendering stage starts (so it gets into the static acceleration structures for best tracing performance.), if one has the ram to fit the resulting displaced mesh.
On one hand, 2D ensures your displacement detail stays even across the mesh even if the original mesh was very haphazard for face areas, while 3d displacement and subdivision will inevitably inherit the properties of the underlying geo (so a subdivided poly which is bigger than another will lead to bigger sub-triangles), but cares naught for the curvature of the underlying mesh.
View-dependancy helps mitigating that a lot, by virtue of an added heuristic condition in the Camera position and frustum, which then is free(er!) of the original topological constraints at subdivision time, but may, when detail is too low, or the setup is slightly off, lead to visible artefacts.
Either way, 2D displacement has very specific uses which are dictated by its technological inception.
I'd concentrate on the type of workflow that these days is common across all other rendering engines: subdivision and displacement of said subdivided vertices, ideally with a 1:1 correspondence between sculpted, and rendered topology.
But then, it's not my job to deliver characters anymore, so take with more than a pinch of salt, and make sure you do your own testing.