I have been wondering if PreciseConvexDecomposition can on occassion provide a better solver for collisions than Default because in the document for it’s release over a year ago it stated (read it here) that:
The use of the word may could suggest that in some scenarios PreciseConvexDecomposition could be either on-par or perhaps even faster than the default solver. I was wondering if anybody has actually done any research or has any tools in which to identify if a union running the two different collision fidelities has higher or lower performance.
In some cases the decomposition geometry in PreciseConvexDecomposition seem to be lower than those in the Default collision fidelity, though notably, it is usually that Default is lower.
I will go ahead and simply mark this as solved for now; since I don’t believe anymore research will turn up. However, if anybody finds anything in the future feel free to revive this thread with new information.
I suck with reading the MicroProfiler (which is where I’d assume you’d find it) but you can clone your UnionOperation ~50 times and set their CollisionFidelity to PreciseConvexDecomposition and then set it to Default and see which one takes longer to compute.
If you don’t have a noticeable freeze, clone your UnionOperation ~100 or ~200 times. Really depends on how much triangles are in the UnionOperation has and how good your computer is.
Besides that, I’m assuming you’re trying to optimize while also keeping decent collision. Have you tried using Hull? Default and PreciseConvexDecomposition geometry are laggier than Hull or Box.
Here’s a helpful image I created demonstrating the different CollisionFidelities. If you want to read more about MeshPart (and UnionOperation) performance, read it here.
If you want to easily see the collision bounds without playtesting, use this plugin by CloneTrooper1019. It’s virus free.
This is a very crude and long-winded method to determining which is more performant, but I’m guessing that sad truth is there is not really any better resources at present for easily determining it.
Thanks, but I’ve already optimised all the collisions for MeshParts and UnionOperations in my game this was more-so abit of side research since it’s not well documented if such a scenario is possible - but I would assume in theory that both solvers have their limitations, upsides and downsides and in some cases the PreciseConvexDecomposition can perform better than the Default solver however rare that may be.
This is a really well done resource for people who might not understand how the basics of 3D modelling works.
This is also a great plugin that I already use for optimising my collisions.
I will go ahead and simply mark this as solved for now; since I don’t believe anymore research will turn up. However, if anybody finds anything in the future feel free to revive this thread with new information.
