On Thu, 21 Oct 2010, Paul Martz wrote:
On 10/21/2010 10:35 AM, John Kelso wrote:
I get similar results for setting serial draw to on or off.
Can you double check this, John? My understanding is that when you have
serialize set to on (the default), then you get a draw thread per window and the
per-thread draw time (from stats display) is X. But they run sequentially, so
the total draw time is X*N, where N is the total number of draw threads. When
you turn serialize off, the per-thread draw time increases to X*N, but the
threads run concurrently so the total draw time doesn't change and is still on
the order of X*N.
So the results for serialize on or off are actually quite different. Am I
understanding things correctly?
--
-Paul Martz Skew Matrix Software
http://www.skew-matrix.com/
John is away this afternoon, so I'll reply.
John's statement is correct, where "similar" is total time(frame rate) is
only slightly different regardless of serial draw on or off. Below is what I
observe when we run the test program. This behavior is true in both the
original program with a single scene graph and the modified test with a
complete scene graph per screen.
OSG_SERIALIZE_DRAW_DISPATCH=ON
With the stats displayed, the draw bar graphs for N threads are nicely
stair stepped as you would expect for serialized draw with a total of X*N
time.
OSG_SERIALIZE_DRAW_DISPATCH=OFF
With the stats displayed, the draw bar graphs for N threads are clearly
parallel and all the same length. The total time is actually X*N+A where A
is some additional time.
So like Robert originally noticed when he implemented
OSG_SERIALIZE_DRAW_DISPATCH, serial draw is faster than parallel draw for a
particular hardware configuration. However, the problem remains - the more
screens (GPUs), the slower the frame rate.
-Steve Satterfield
_______________________________________________
osg-users mailing list
[email protected]
http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
