|
 |
Hi there.
Here're the results from the tests I promised. Sorry it took me so
long, but it's media, you know...
Please see p.b.s-f for the code and p.b.i for example images.
Type 1 - multiple densities
type1_1.tga, Quality 2 - 7 minutes
type1_2.tga, Quality 3 - 21 minutes
type1_3.tga, Quality 4 - 29 minutes
Type 2 - multiple containers
type2_0.tga, Quality 1 - 4 minutes
type2_1.tga, Quality 2 - 6 minutes
type2_2.tga, Quality 3 - 17 minutes
type2_3.tga, Quality 4 - n.a. sorry, dad played freecell :(
type2_2.tga, Quality 5 - 37 minutes
Multiple containers are faster and with better quality. The examples
posted in p.b.i. show a particle system of 50 particles. Using
multiple containers, the result was much less grainy and was rendered
in 6 minutes vs.29 for multiple media. The results differ slightly but
that's due to the small accuracy of type 1.
I also did a 100 frames 100x100 anim. It started with a single
particle and added a particle every frame (if anyone's interested I'll
post it). Here're the stats:
Type 1 (multiple media) statistics:
Total time: 3:16:34, render 3:16:11, parse 0:0:23
Intersections:
Shape Tests Succeeded Percentage
Sphere 419435 1708050 70.60
Bounding Box 3831950 3830350 99.95
Vista 1030320 1030320 100.00
Media:
Intervals 4963140
Samples 34746921 7.00 per interval
Transmitted rays 1418040
I-Stack overflows 7125
Memory:
Smallest 26
Largest 8216
Peak 134132
command line:
+w100 +h100 +mb0 +uv +kfi0 +kff99 -d +a0.4 +r2
Type 2 (multiple containers) statistics:
Total time: 2:59:18, render 2:58:39, parse 0:0:16
Intersections:
Shape Tests Succeeded Percentage
Sphere 54614525 37059270 67.86
Bounding Box 266093261 113576463 42.68
Vista 19936209 12944728 64.93
Media:
Intervals 45169600
Samples 313641496 6.94 per interval
Transmitted rays 7248496
I-Stack overflows 989
Memory:
Smallest 26
Largest 8216
Peak 198691
System: K6/233 96 RAM running POV-Ray for windows 3.1e on 98
command line: +w100 +h100 +mb0 +uv +kfi0 +kff99 -d +a0.4 +r2
(These look fine using Lucida Console)
The main disadvantage of type 1 is that all densities are calculated
for every interval and every sample. This leads to graininess and
inaccuracy because small radius densities may not be hit at all.
The main disadvantage of type 2 and it is that max_trace_level and
max_intersections should be really high. This could be really painful
in scenes with reflective / refractive objects. I've put in a type 3
which uses merge to avoid the problem, but preliminary tests show that
it's slower than type 1 (though more accurate).
Well, that's it for now. Any comments?
Peter Popov
ICQ: 15002700
Post a reply to this message
|
|
