Mesh resolution

[magilla]

I haven't noticed in all the tests I have done but does the simulation run faster with lower res mesh for collisions or does it not affect the sim time?

[Ivaylo Katev]

yes, if the mesh has too big triangle count (1M or more), it's possible that significant part of the simulation is spent in the voxelization. this is especially true when the geometry is moving, because the static geometry is scanned only twice, but the moving geometry is scanned at every step.

[Neilg]

Does that mean that if you have a million poly mesh, but it's static you only see slowdown on the first frame or whatever and the rest go by as usual?

I was about to start modeling a low poly collision mesh and it'd be great if i dont have to...

[Ivaylo Katev]

yep, exactly. you can see the slowdown the first two frames and then the sim runs significantly faster.
however, the scan is performed not only when the geometry is moving, but also when the geometry is a source and the source params are animated. if you geometry is static and is not an animated source, then you can use hi-poly model, the slowdown in the beginning is negligible compared to the entire simulation

[Neilg]

awesome news, thank you!

[joshclos]

Continuing this - does phoenix automatically detect if geometry is static or is that based on the per object "Animated Vertices" property? So if the geometry is static but "Animated Vertices" is on by default, will it continuously calculate the voxelization until the option is disabled?

[Svetlin.Nikolov]

Ah, no - if the geometry doesn't do anything, there will be very small overhead compared to when the geometry is actually transforming or deforming. The Animated Vertices property just completely turns off the processing of deforming vertices, and this will result in a bit less memory used - if you are colliding the simulation against huge meshes, this would make a difference, but simulating with such geometries would be painfully slow anyway, and furthermore the larger the geometry is, the higher chance is that there would be issues with it like inverted normals, overlapping faces etc that might make the solver blow up, so in a best case scenario you shouldn't be using such geometries anyway...