Using an SDF Mesh collider makes my simulation run like 20x slower than using convex hull. I’ve set the SDF Resolution as low as I can (1 in many cases) without getting Tensor errors.
Any tips on making the SDF Mesh more performant? Should the performance be better if two meshes are not close to each other? Or should the performance be constant regardless of proximity-to-collision?
Hi, do you have an usd file such that we can reproduce the performance problem?
About the SDF performance: The SDF grid resolution is usually not the performance limiting factor, it mainly affects the memory consumption of the SDF. For performance, the number of triangles in the mesh is relevant. I cannot say how complex the mesh is. It is possible to have the rigid body API on an XForm which has two meshes as child nodes: One for rendering with as many triangles as you like and one for collision with the collider api. The one for collision can be hidden such that it won’t get rendered. For the collision mesh you can reduce the number of triangles in an external tool such that it matches the shape of the original mesh still very closely but contains a lot less triangles. Convexes will always be faster since the usually only have a maximum of 64 faces. A triangle mesh can have millions of triangles and each of them needs to get tested (the process is accelerated but still much more expensive than for a simple convex).
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.