|
 |
Vaclav Cermak <dis### [at] disnel com> wrote:
> Great, I'am eager fot it ;-). But I have strong feeling, that theese
> 300% slower scenes are my test scenes with meshes ... Right?
Well, yes and no.
The bad news is that yes, the 300% scene (+200% of original CPU time) is one of
yours :P
The good news is that no, actually all your *original* scenes happen to be
*boosted* in speed, ranging from a "lousy" -10% (rad_test04) to a friggin' -75%
(rad_test03) difference in CPU time (as compared to MegaPOV 1.2.1, which is the
best-optimized version of POV 3.6 I have available for my AMD64 Linux machine).
For some yet unexplained freakish reason, it's just the experimental versions
you produced later that still suffer compared to 3.6: A moderate +20% for the
flat-mesh version of rad_test04, and those nasty +200% for the recursion depth
1 variant.
(Note that I wasn't able to test your rad_test01 scene.)
The only other scenes that suffered noticeably are the "patio-radio" sample
scene using radiosity settings variant #4 (+25%), the "rad_def_test" sample
scene using OutdoorLQ settings (+10%, but that's just about 1 second), the
"radiosity" sample scene using settings variant #4 (+15%), and one of my own
scenes with a very pathological geometry (some people may recall that scene
with the almost-but-not-quite closed door; +150%; doesn't render properly
anyway, regardless of version).
Ah, yes, and the "balcony" sample scene suffers +20%, and an occasional segfault
for some reason still not fully understood.
Note that we're talking about CPU time here; on a QuadCore system running full
throttle, SMP turns even the worst of these cases into winners when it comes to
wall clock time.
BTW, thanks for providing your scenes. You may like to hear that they proved a
particularly valuable resource for radiosity testing and tuning - not only as a
speed benchmark, but also for comparison of quality, as the scenes exhibited
stong artifact responses to certain tweaking, revealing severe issues with
those settings I would not have been able to see otherwise.
Post a reply to this message
|
|
