I am, glad finding useful I haven't run comparison, neither I had plan to. To be honest, not sure how I would test raycasting of physX vs ECS octree based approach. I suspect method would be, to to place n * 10 thousands of objects in the scene, for each approach, and test raycast then. In my case, each octree is static structure stored in own entity. Meaning octree physically never moves, nor rotates. Each octree entity, has buffer arrays, holding structure of nodes. Every node, which is always AABB, stores reference to entity as an option. Technically, don't have to be entity reference. Can be any arbitrary index, to something you like. I.e. to array. I would suspect, some similar approach is taken with physX. Then testing OBB vs AABB Only thing is moving, is ray for casting, or collision test boundary. But in this case, these referenced entities, has mesh component and transform, allowing for rendering. Then I take ray cast from mouse, and translate its position and rotation (as per last video), relative to the entity (s) mesh. That relative rotation of ray, is actually applied for the octree. To be honest, I am not familiar with MultiBox. But yes, I haven't been optimizing it for boxes / mesh collisions. Providing I got term right, CSG (Constructive Solid Geometry) probably is better fit, for voxel structures is such case.