Hi Daniel,
Daniel Wickeroth wrote:
> Hi,
>
> I just did some tests on OpenSG and OpenSceneGraph to find out which
> better suits my needs (Performance) So I wrote some little testing
> programs, loaded a VRML or an OBJ file (OpenSceneGraph doesn't seem to
> support VRML) and measured the framerates.
>
> You can find the results here:
> http://www.uni-koblenz.de/~schmecko/Seite.htm
Could you put the programs and the models up there, too? That would
allow me to try to reproduce it and see what's going on.
> As a basic file I used the tie.wrl that comes with OpenSG. To create
> larger scenes I simply multiplied the number of the tie-fighters in
> the scene (up to 729= 9*9*9). Since OpenSceneGraph doesn't seem to
> support VRML I converted the files into the .obj-format. I tested both
> (wrl and obj) with OpenSG.
>
> Some things were pretty strange.
> First of all OpenSceneGraph was sometimes twice as fast as OpenSG
> (tie243.wrl), which I did not expect. I tried to match the setups as
> closely as possible given the short time I had. But even if there was
> a slight advantage for OpenSceneGraph, the results seem odd to me.
Looking at the framerates it looks like you have sync to vblank enabled.
That is generally a bad idea for benchmarking, as small differences in
actual speed can have a very big effect on framerate. Try to rerun the
tests with that turned off.
I'm surprised about the loading times for obj. That looks quadratic to
me, which I wouldn't have expected. If you could put the models up that
would be helpful.
> Secondly, OpenSG 1.8 seems to be slower than OpenSG 1.6. But this I
> cannot really say, because with OpenSG 1.6 I only built my code
> myself, and used the precompiled binaries from the web. With OpenSG
> 1.8 I compiled everything myself, inlcuding the OpenSG libraries, with
> the VS 8.0 compiler.
Just to rule out simple things: I assume you used optimized libs for
everything?
> I would be glad to get some comments on the tests. Maybe some
> improvements on the procedure? If you want I will gladly upload some
> sceenshots and more info about the testing procedure.
The tie is a pretty small model, i.e. few polygons/node, which leads to
putting more weight on traversal than rendering. In that arena I'm not
very surprised OSG has an advantage, as it has a simpler structure. I
wouldn't expect a factor of two, but that might be caused by the vblank
issue.
I would like to reproduce the tests, so if you could put your code and
models somewhere, I'd appreciate it. Note that I'm at VR in Charlotte,
so I won't be able to look at it before next Thursday, and even that is
tight because I'm leaving again Friday at noon (busy times... ;).
Yours
Dirk
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
_______________________________________________
Opensg-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/opensg-users
