[feynman]

For detailed displacement (a stucco frame on flat wall/ceiling ) is it better to output a highly subdivided surface from one's 3D modelling application, or is it better to use a high subdivision factor in Studio?

With 32MB RAM, I cannot subdivide a plane with Pretessellated more than 1024 - if I do, the PC runs out of memory and at 1024 the displacement lacks much detail. So, what's best practice?

Tack!

[JDHill]

Apply a non-displacement material to the full plane; simultaneously, cut a copy of the plane very closely to the displaced area, apply to it a displacement material, move it back (behind the full plane) a bit, and let it displace through the full plane. The point being: avoid subdividing geometry that will not actually be displaced.

[Mihai]

1024 subdivisions? Imagine if your initial plane has two triangles. 2 x 2 x 2 x 2......do that 1024 times and figure out how many triangles you end up with

How close are you looking at these displacements? It seems pretty close if even 1000 thousand subdivisions doesn't give you what you want. Have you tried on the fly in this case? And yes, no need to subdivide that entire surface if only 5 % of it needs displacement. It may also be the case that the displacement texture is not high rez enough. If each triangle in the geometry has already reached the detail of a single pixel in your texture - more subdivisions won't give you more detail. So be careful also how you UV map this so the mapped area doesn't go to waste. If you've mapped this as a single UV map across a big plane, then that intricate detail in the corners will only get a few pixels worth of displacement info.

[seghier]

maxwell subdivide all the geomerty or only the part need displacement ?

[feynman]

Apply a non-displacement material to the full plane; simultaneously, cut a copy of the plane very closely to the displaced area, apply to it a displacement material, move it back (behind the full plane) a bit, and let it displace through the full plane. The point being: avoid subdividing geometry that will not actually be displaced.

Once again, a JDHill suggestion does the trick.

I was under the impression that Maxwell tessellates adaptively (no extra triangles generated by subdivision where none are needed).

[feynman]

1024 subdivisions? Imagine if your initial plane has two triangles. 2 x 2 x 2 x 2......do that 1024 times and figure out how many triangles you end up with

How close are you looking at these displacements? It seems pretty close if even 1000 thousand subdivisions doesn't give you what you want. Have you tried on the fly in this case? And yes, no need to subdivide that entire surface if only 5 % of it needs displacement. It may also be the case that the displacement texture is not high rez enough. If each triangle in the geometry has already reached the detail of a single pixel in your texture - more subdivisions won't give you more detail. So be careful also how you UV map this so the mapped area doesn't go to waste. If you've mapped this as a single UV map across a big plane, then that intricate detail in the corners will only get a few pixels worth of displacement info.

It looked too coarse with less than 1024, and even 1024 was not sufficient. JDHill's suggestion makes sense. I had wrongly assumed that the subdivisions are created adaptively, in my case that would have meant that not many triangles are created in the flat 255/255/255 area where no displacement detail exists, similar to a Dirichlet tessellation or how a TIF compressor compresses areas where no change in pixels occurs.

Would be a good idea to include JDHill's suggestion in the Knowledgebase or Documentation - it may be obvious to visualisers and animators, but not so obvious to design engineers or product developers.