We've actually improved the performance so much that we've canned occlusion queries and just keep normal per-node frustum culling (exception is for instanced stuff where we do GPU-based frustum culling).Rendering large models, like 1GB meshes of complete cars seems to just load the GPU so much. I guess its still rendering all the sub-meshes even if they are not finally visible.
We have plans to do per-meshbuffer culling very soon.
We've done amazing changes to the ways that assets are loaded (meshes and images) as well as how you create GPU objects from CPU data.So you made a lot of change to Irrlicht 8.4 for parallel processing then? Does it also handle sub-mesh rendering so as not to render unless visible?
Its quite hard to explain without writing pages of A4 (which has already been written in an open google doc linked to on FB and in my thread)