Greg M. Johnson nous apporta ses lumieres en ce 21-01-2007 08:03:
> I'm trying to duplicate a complicated floor pattern I see in a photo. It's
> texture has texture_map's couple of layers deep.

> I have one scene in an animation that is rendering unusually slow compared
> to some of the others.  In considering the suspects I'm wondering if it's
> the floor tile!

> Q: do multi-layered patterned textures slow up render speed?

> Any tips on how to speed it up: 
>         is it better stick to patterned pigments and do without "normal" in the
> texture?
>         are there any problems with the way in which you define a pattern--is there
> a trick to defining it all in one gulp?

Layered textures are slower than simple oned as each layer have to be evaluated 
(but only if behind a transparent part of the preceding one(s)). So, yess, 
layered textures do slow you down, BUT there are many other things in most 
scenes that can have much more effect.
Do you have refflections? What is been reflected?
Do you have transparent or partly transparent object? With an ior? Using 
dispersion? Those will slow you down.
In almost all scenes, normal have only a negligible effect on speed, exeption: 
in a radiosity scene with "normals on" set in the radiosity block = big slowdown.
Look at the geometrys in your scene, do you use many difference and 
intersections? With LOTS of small parts removed? This is a common source of 
slowdowns.
A trick is to turn on the histogram feature and look at the result. Bright parts 
= slow parts. Use the "+htn" switch on the command line to turn on and set the 
output to PNG. Using that switch, there is a 1 to 1 relation between the final 
image and the histogram output.
On the windows version, there is a "command line" box next to the resolution 
selector.

Some tips:
Take a look at the scene statistics, do you have any entries with low percents?
Try removing objects that you don't see. If some object is only seen in 
reflections and the reflection is small or distorted, replace it with a mockup 
that have about the same apearance.
Do you have differences where you remove LOTS of small parts? When evaluating 
such an object, you need to check for ALL the removed parts for each pixel: all 
the bounding box overlap! It may be faster to put those small removed parts in a 
merge. That way, all components are treated as a single object having only one 
common bounding box.
Do you use any sphere_sweep? Are they long and thin? With broad curves? If so, 
try splitting them into a collection of shorter sweeps.

-- 
Alain
-------------------------------------------------
WARNING: the crumsumpten of alcohol may Mack you tink you kan tpye reel gode

Post a reply to this message