koobas.hobune.stream
Does combining meshes yield a net performance gain?
I recently completed a small tool to make combining meshes easier. After using it to combine the static contents of a room in a test scene, I saw the following:
The draw calls were reduced because submeshes using the same materials were combined
The tri count was increased by 30% (mostly from shadow maps)
The number of shadow casters was reduced by 30%
How do I determine if the outputted combined mesh will yield a net performance gain given these 3 changes?
I'm looking for something more accurate and predictive than that. If I'm observing an asset in isolation, the FPS is too high to show any perceivable difference.
I could populate a scene with hundreds of copies and see what difference is perceived then, but that's an unwieldy approach.
What I really want is to have a better grasp on how to interpret tris vs. SetPass calls vs. Shadow Casters to be able to logically conclude which of the merged and unmerged versions performs better.
I don't think you'll get a decent answer, or for sure not a short one. From my somewhat limited experience it varies by platform and even device, and of course your scene and other code/assets. Not trying to discourage, just saying I wouldn't wait around holding my breath for an answer.. I'd just dive in and try both.
