Hi All,

I was curious to see how 3.6 performance compares to 3.5, particularly
considering the team was focussed on internal organisation of the program in
prep
for the major rewrite to come in 4.0.

So here are my benchmark results.

Platform:
OS:  WinXP Home
MoBo:  ASUS A7N8X Deluxe Ver 2.0
CPU:  Athlon XP 3200+, FSB @ 200MHz, Freq. Mult 11.0x (2200MHz)
RAM:  1024MB Corsair TwinX 3200 LL (Low-Latency) DDR @ 200MHz (a.k.a. "400"
or "PC3200")
[This RAM comes as a matched pair of 512MB sticks plugged into opposing
banks for performance]
RAM Latency:
  CAS: 2
  RAS to CAS delay: 3
  RAS precharge delay: 2
  Active precharge delay: 6
LAN disabled during test to eliminate network processing overhead


Reported by Pov-Ray for Windows 3.5

Kernel CPU:  0.38 sec
User CPU:  1905.83 sec
Total CPU:  1906.20 sec

Parse:  1 sec
Photon:  49 sec
Trace:  1871 sec (31 min 17 sec)
Total:  1921 sec (32 min 1 sec)

Render averaged 77.36 pps over 147456 pixels


Reported by Pov-Ray for Windows 3.6

Kernel CPU:  1.38 sec
User CPU:  1670.45 sec
Total CPU:  1671.83 sec

Parse:  1 sec
Photon:  40 sec
Render:  27 min 18 sec (1638 sec)
Total:  27 min 59 sec (1679 sec)

Render averaged 88.20 pps over 147456 pixels

So that's a saving of four minutes or about 14% improvement compared to Ver
3.5.  The team has done well  :-)  I notice the formatting of the timing
statistics is a little different also.


Cheers,
  Brian

Post a reply to this message

Brian Elliott <bel### [at] gilcomau> wrote:
> So that's a saving of four minutes or about 14% improvement compared to Ver
> 3.5.

  If I'm not completely mistaken, this is due to the optimizations made
to the photons code.

  In a scene not using photons there shouldn't be such a difference.

-- 
plane{-x+y,-1pigment{bozo color_map{[0rgb x][1rgb x+y]}turbulence 1}}
sphere{0,2pigment{rgbt 1}interior{media{emission 1density{spherical
density_map{[0rgb 0][.5rgb<1,.5>][1rgb 1]}turbulence.9}}}scale
<1,1,3>hollow}text{ttf"timrom""Warp".1,0translate<-1,-.1,2>}//  - Warp -

Post a reply to this message

"Warp" <war### [at] tagpovrayorg> wrote in message
news:40ced0cf@news.povray.org...
> Brian Elliott <bel### [at] gilcomau> wrote:
> > So that's a saving of four minutes or about 14% improvement compared to
Ver
> > 3.5.
>
>   If I'm not completely mistaken, this is due to the optimizations made
> to the photons code.
>
>   In a scene not using photons there shouldn't be such a difference.
>
> -- 
> plane{-x+y,-1pigment{bozo color_map{[0rgb x][1rgb x+y]}turbulence 1}}
> sphere{0,2pigment{rgbt 1}interior{media{emission 1density{spherical
> density_map{[0rgb 0][.5rgb<1,.5>][1rgb 1]}turbulence.9}}}scale
> <1,1,3>hollow}text{ttf"timrom""Warp".1,0translate<-1,-.1,2>}//  - Warp -


I did know the photon code is improved.  I could also read the release notes
for more overall information (which I haven't done yet).  I had simply
assumed that the photon code improvements would only manifest in the photon
calculation stage, but that only accounts for nine seconds difference in the
two runs, not the whole four minutes.  But then I conveniently forgot:  the
engine also has to read the photon map during rendering, perhaps that was
improved as well.

I'll compare a couple of scenes without photons, just to see for myself if
there are any performance differences beyond that.  This is all probably
splitting hairs anyway. But I was curious and then surprised because I
expected to find negligible change -- but 14% is significant.

Brian

Post a reply to this message

Brian Elliott schrieb:

> I'll compare a couple of scenes without photons, just to see for myself if
> there are any performance differences beyond that.  This is all probably
> splitting hairs anyway. But I was curious and then surprised because I
> expected to find negligible change -- but 14% is significant.

I think a lot of improvement come's from the compiler used. If the 
compiler used for 3.5 is another compiler than that used for 3.6 there 
can be a big improvement.

-- 

mat### [at] matweide
http://www.matwei.de

Post a reply to this message

=?ISO-8859-1?Q?Matthias_Wei=DFer?= <mat### [at] matweide> wrote:
> Brian Elliott schrieb:
>
> > I'll compare a couple of scenes without photons, just to see for myself if
> > there are any performance differences beyond that.  This is all probably
> > splitting hairs anyway. But I was curious and then surprised because I
> > expected to find negligible change -- but 14% is significant.
>
> I think a lot of improvement come's from the compiler used. If the
> compiler used for 3.5 is another compiler than that used for 3.6 there
> can be a big improvement.

Speaking of which, I tried compiling with GCC 3.4 using the flags
'-mach=ahtlon-xp -mtune=athlon-xp -mfpmath=sse' and saw around 10% speed
improvement.  Without changing one line of code, simd can still benefit a
double-precision app like POV :)

....Chambers

Post a reply to this message