Christoph Fasching

Hi all,

In Houdini 22.0.466 I’m merging four RBD setups (shingles, drywall, stone, wood): 18k pieces and 68k Glue constraints. Every time I change the Merge input order, I get a different simulation. Each result looks plausible, but they’re clearly not the same sim.

I checked the solver inputs per piece name and per constraint pair. Apart from element order, they’re identical between orders. Names are unique, there’s no anchor_id, and there are no detail attributes.

I attached two flipbooks: same setup, different merge order. If I keep the merge order and just reset the sim, the results match exactly.

Is it expected that Bullet gives different results purely from element order?
What’s the best way to make it reproducible (sort by name, Randomize Constraint Order, more iterations)?

Happy to share a .hip. Thanks!

What is your voxel size on the rasterize node?
Your wrangle sets pscale between 0.01 and 0.02.
If the voxel size on the rasterize node is about the same or larger, each particle splats into only one or two voxels and since voxels are arranged in a grid they snap to that grid and you get stepping.
BUT
even then, it might just mostly be a viewport issue. Try rendering it with karma or redshift or whatever your renderer of choice and see if you can replicate the stepping there.

Drewzee
Not to dog pile but I'm starting to run into some quirks with X11 on Pop OS 22.04 (its basically ubuntu). So I'm adding a +1 for Wayland. Pop OS is so much better than Windows though. Zero crashes in COPs on 22.0.429. Windows crashed constantly, especially when screen recording. But without wayland screen recording is still really really bad/choppy.

What quirks are you experiencing?
I am also on PopOS! 22.04 here but mostly smooth.
Is it specifically with screen recording?