Hi Robert, > Another possibility I missed is the use of doubles in osg::Plane by > default, whereas 1.2 used float by default in osg::Plane.
thanks for the hint. However, I think the problem could be related to the compile/apply of GL objects instead. If I disable the GLObjectsVisitor that my app runs after loading a new scene, both builds of the application start in similar conditions with little difference in memory consumption. At this point I start moving the camera around, and display lists and textures get compiled as they are displayed for the first time. Here the difference between the OSG-1.2-based build and the newer one becomes clear: the newer one eats much more memory, roughly twice as much, while the amount of data loaded (geometry+imagery) is the same. All textures are DXT-compressed, and they come with mipmaps already stored in the image file (DDS). In order to minimize the influence of other optimizations, I've turned off the UnRefImageAfter....etc. flag on all textures. Except for the different version of OpenThreads and OpenSceneGraph, both builds share the same code and 3rd-party libraries. Any other clues? Thanks, Marco _______________________________________________ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org