Hi Kirill, This version of your scene graph is clearly GPU limited, so batching everything into a single geometry certainly avoids CPU bottlenecks, but... it probably not ideal as it's far from a balanced scene graph, and the batching is now far too coarse grained.
Ideally you should have your scene graph so it forms a balanced quad tree, with the leaves on the quad tree providing the geometry, be it instanced, shared or just merged as you've done here but at a far more fine grained way. So rather than one geometry with 6million vertices you should perhaps aim for 10,000 vertices per Geometry all ground grouped spatially so that we you build a quad tree above it it's balanced. This combination will leads with good culling performance and reduce chances of cache misses on the GPU causing big stalls. With very specific types of scene graph usage like yours it may well be best to create the scene graph structure yourself as trying to use general purpose off the shelf optimizers to try and fix your scene graph won't ever be as good as optimizers written specific for your specific usage case. Robert.
_______________________________________________ osg-users mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
