[osg-users] Chunked lod

Robert Osfield robert.osfield at gmail.com
Mon Jun 25 01:23:33 PDT 2018

Hi Tyler,

Modern graphics cards love coarse grained chunks of static geometry,
they can churn through at millions of vertices a second without
breaking sweat.

Once you start make the scene graph more fine grained or making more
of the geometry dynamic so that it has to be updated per frame then
the more your application will be CPU limited and won't be able to
make the most of incredible throughput capability of the graphics

Twenty+ years ago the speed different between GPU and CPU's was much
lower so it was more beneficial to avoid sending geometry to the GPU
by using techniques like ChunkedLOD, these days the speed ratio is
quite different, the GPU have got faster much more rapidly than CPUs,
so techniques that were really useful previously now aren't optimal.

With modern hardware if you did want to do some form of continuous LOD
then you would be best to look at pushing more work at the GPU and
keeping as much load off the CPU as you can.


More information about the osg-users mailing list