Re: Quesa - "Most Wanted" list

[James W. W. <os...@jw...>]

On Jun 2, 2004, at 6:38 AM, Joseph J. Strout wrote:

>>> 4. Improved speed -- though this has improved recently, we still 
>>> find that object performance degrades pretty quickly for scenes with 
>>> lots of objects.  (Some of this may actually be in the REALbasic 
>>> layer on top of Quesa, but we'd like to at least look into where the 
>>> current bottlenecks are and see what can be done about them.)
>>
>> "degrades pretty quickly" meaning as more than a linear function of 
>> the number of triangles?  Have you tried to separate out the effects 
>> of lots of triangles, lots of textures, and lots of groups?
>
> Yes; the issue is lots of objects (whether this is because it's lots 
> of groups, or because it's lots of TriMeshes, is unclear), not the 
> number of triangles.  A scene with 4000 triangles in one TriMesh 
> renders very quickly; the same scene composed of 400 objects 
> containing 10 triangles each renders poorly.

I did an experiment that illustrates this pretty well.  I started with 
the Multibox test in Geom Test, 1000 boxes, each of which is cached as 
a group containing 6 Trimeshes.  On my Powerbook, it runs at about 15.2 
FPS.   Then I decomposed the boxes and flattened the hierarchy, so that 
all 6000 TriMeshes belonged to a single group.  That brought it up to 
20.4 FPS.  Then I merged TriMeshes with like attributes, resulting in 
just 6 TriMeshes, and that ran at 135 FPS.

In case anyone wants to play with these, I saved them as:

<ftp://ftp.jwwalker.com/multibox-orig.3dmf.zip>
<ftp://ftp.jwwalker.com/multibox-decomposed.3dmf.zip>
<ftp://ftp.jwwalker.com/multibox-merged.3dmf.zip>
--
<http://www.jwwalker.com/>