|
![](/i/fill.gif) |
Am 22.02.2012 06:29, schrieb taftj:
> I am rendering at 1024x768 with AA threshold 0.0, AM2, R5, and Q11
That will cause 4^5 = 1024 (!) rays shot /unconditionally/ for each and
every pixel, even in uniformly colored areas where a single ray would
theoretically suffice. Even with an exceptionally low threshold like,
say, 0.01, I would expect a big win in render speed. Most people go with
the default threshold of 0.3, which I'd expect to win you a factor of
100 or even more. You can also usually get away with a lower recursion
depth; R3 is quite common.
> light_source { 2*<5, 5, -15>*1 color White*1.5 area_light<5, 0, 0>,<0, 0,
> 5>, 10, 10}
Similar here: You'll get 10*10 = 100 rays shot /unconditionally/ per
shadow test (which /multiplies/ by the aforementioned factor due to
anti-aliasing); try the "adaptive" keyword, usually "adaptive 1" is a
good start and will cause only (2^1+1)^2 = 3^2 = 9 rays to be shot for
fully lit or fully shadowed regions, and even for the umbra it will be
more like 20 to 30 samples due to the recursively adaptive algorithm);
that might win you a factor of maybe 8. (Note that when using the
"adaptive" keyword, it is recommended to have an area light of (2^N+1)
by (2^M+1), e.g. 9 by 9, which is actually a quite common value.)
Sphere sweeps are comparatively challenging to render, and your scene
has about 50 of them (one with about 50 control points); you may want to
try different approaches to avoid them altogehter. Using SDL to compute
corresponding meshes might also be an alternative.
Post a reply to this message
|
![](/i/fill.gif) |