In theory all ninite does is mesh streaming.
Meaning a mesh which is round and has 3m tris, can be displayed with 1.5m tris since about 50% of it is not always visible.
Further, it should recompute geometry so that only the amount needed for the distance at which the item is viewed is used (this what epic claims, regardless of how well they try to do this, this is never going to be true). So insted of a 50% only reduction, you may get up to a 90% reduction if you are further away from the item.
Obviously that isn’t going to help with meshes like grass, which are normally flat/2d panels, and need to be rendered as a whole (with proper LODs probably).
The heightfiled meshes would probably work better to do terrain(gravel) displacement at runtime.
It’s probably useless overhead compared to just creating a proper mesh and levaraging parallax materials.
Still don’t really see how procedural tessellation would help you. Or anyone for that matter when it comes to stuff like this.
If you have a very low base tris object, and you split all the tris vectors / generate 4 triangles out of every traingle you have more resolution.
Unless something modifies the position of the extra geometry the result is visually identical, just 4 times more expensive in tris count.
When you do modify the position of the geometry, you still won’t get the results you expect 9 out of 10 times:
Remeber that the approach is also procedural.
Say you have a mesh with 1mm x 1mm x 1mm tris. You just split that into 4 .05mm sided tris.
What good is that going to do / who’s ever going to see it?
No one = wasted computing power.
Had they done this right originally, you could set a minimum value for tessellation to have effect, but they didn’t do this right, and its always been iffy at best.
Where does it really hurt?
Landscapes.
Compute power with tessellation on is extremely heavy on LOD transitions (because the whole landscape system is trash, but that’s besides the point).
Where can it be useful?
On flat geometry to which you want to attempt to add some layer of shape - rock walls or similar.
Where is it useless?
Skeletal meshes. Spline meshes. Really anytning which is already intended to have its vertex altered by something in the engine. Be that at runtime or during the creating process.
Thats pretty much all there is to it, and why it was removed.
It serves no real purpose anymore, and was never done right in this sorry excuse of an engine, so it’s easier to yank it completely than it is to continue allowing users (we should say amateurs) to make bad uses of it.
And before anyone brings up snow trails.
No.
You make specific meshes with the right amount of tris for that. Similar to water. In a way that doesnt cause seams when you apply LOD reductions. Or you use the heightfiled mesh thing, which works ok in a pinch for stuff like snow.
(Neither approach is ideal, having an analytical mesh deform which uses more or less triangulation depending on angualr incidence would outperform any system in place this far).
