Hi Robert,

You won't see an increased framerate for applications that are already 
GPU-limited. The easiest way to see an improvement is to simulate an expensive 
cull traversal, see the example in the first post. The OSG cull traversal may 
not be so expensive by default for a balanced scene graph, but cull callbacks 
are often used to update animations, lighting or other user tasks so I would 
consider it important that filling the cull stage with more work scales well. 
You could also try adding more cameras to see how that scales. Generally I 
found that in CullThreadPerCamera threading mode without this patch, you very 
quickly become CPU limited because only the last camera's culling stage can 
overlap with the rendering thread. DrawThreadPerCamera remedies the problem, 
but can't spread the culling work across CPU cores.

I guess we have increased latency now because with an earlier frame break, the 
update and culling threads can jump ahead of the rendering thread further 
causing more work to be queued up. Unsure how to resolve this. Most users would 
prefer higher framerates over a small decrease in latency. But then again, 
higher framerates aren't a given, only if you were previously CPU limited.

It's a low priority request anyway - for most OSG users the culling stage will 
have negligible impact. I am unsure how to fix the increased latency issue, so 
maybe we should drop the idea altogether.

Cheers,
Jannik

------------------
Read this topic online here:
http://forum.openscenegraph.org/viewtopic.php?p=65222#65222





_______________________________________________
osg-submissions mailing list
[email protected]
http://lists.openscenegraph.org/listinfo.cgi/osg-submissions-openscenegraph.org

Reply via email to