First, my system configuration:

  Machine Model: Power Mac G5
  CPU Type: PowerPC 970  (2.2)
  Number Of CPUs: 2
  CPU Speed: 2 GHz
  L2 Cache (per CPU): 512 KB
  Memory: 2.5 GB DDR SDRAM
  Bus Speed: 1 GHz
  Boot ROM Version: 5.07f0
  System Version: Mac OS X 10.3.1 (7C107)
  Kernel Version: Darwin 7.0.0
  Boot Volume: Macintosh HD

Now, the benchmarks.

I ran the simple benchmark from the Mac POV-Ray menu, which runs the
benchmark with the following options:

Input file: Macintosh HD:private:tmp:501:TemporaryItems:benchmark.pov
(compatible to version 3.51)
  Remove bounds........On   Split unions.........Off
  Library paths:
    Macintosh HD:Applications:POV-Ray 3.5:include:
Output Options
  Image resolution 100 by 100 (rows 0 to 100, columns 0 to 100).
  Output file: benchmark.tga, 24 bpp Targa
  Graphic display.....Off
  Mosaic preview.......Off
  CPU usage histogram.Off
  Continued trace......Off
Tracing Options
  Quality:  9
  Bounding boxes......On  Bounding threshold: 25
  Light Buffer.........On   Vista Buffer.........On   Draw Vista
Buffer....Off
  Antialiasing........Off
  Radiosity............Off
Animation Options
  Clock value....   0.000  (Animation off)

I tried all three rendering engines and several different performance option
combinations. I was unable to use the "Fast Functions" performance option
with the G5 engine which resulted in significantly slower photon times (the
scene parsing aborted with an "Internal error" message which I've already
described in detail in the "OS X 10.2" thread).

Here are the results, which were obtained by restarting OS X before each run
and ensuring that no other applications (other than minimal system
functions) were running. All tests were performed three times, and the
average time is shown here:

(Note: All times are expressed in seconds as parsing/photon/rendering/total
times.)

G5 rendering engine:
All performance options EXCEPT "Fast Functions":   4/51/47/102
No "Fast Preview" or "Fast Functions:              4/50/48/102
No "Fast Preview, "Fast Functions",
    or "Background Rendering:                      4/59/55/118
No "Fast Functions" or "Use Temp Memory":          4/51/47/102

G4 rendering engine:
All performance options selected:                  4/34/43/81
All performance options EXCEPT "Fast Functions":   4/49/47/100
All performance options EXCEPT "Use Temp Memory":  4/34/43/81

Standard 3.51 rendering engine:
All performance options selected:                  4/36/45/85
All performance options EXCEPT "Fast Functions":   4/51/47/102
All performance options EXCEPT "Use Temp Memory":  4/36/45/85

From these unscientific tests, it appears that the G5 optimizations actually
hurt performance... the G4 engine was slightly faster than the G5 engine
and the standard 3.51 engine on the test that is directly comparable across
all three engines (the runs where "Fast Functions" were disabled). With
"Fast Functions" enabled, the G4 engine was slightly faster than the
standard engine in both photon and rendering times observed.

I next ran the benchmark using the standard benchmark.ini on the povray.org
website. I won't repeat those settings here; you can get the lengthy INI
file and see them for yourself. Here are the results, averaged over three
runs each and performed after starting OS X fresh and with no other apps
running:

G5 rendering engine:
All performance options EXCEPT "Fast Functions":   4/60/1740/1804

G4 rendering engine:
All performance options selected:                  4/38/1665/1707
All performance options EXCEPT "Fast Functions":   4/59/1735/1798

(I didn't run the standard engine on the long benchmarks.)

These times are pretty good, but it looks like the G4 engine is still the
best optimization to use for now. Hopefully some further tweaking to the G5
compiler options will eke out some extra performance.

Post a reply to this message