Will a New Year bring a New World Order?

This render took 6.8GB, mostly from the high resolution textures used on the
characters. I experimented with both v3.6 and v3.7. The render took about 7h50m
on v3.6. Surprisingly, the render took about 40 minutes with v3.7! On my Quad
Core, I was only expecting a 4X increase, but it looks like there are some
other cool optimizations in v3.7. There are a few radiosity artifacts in the
3.7 version of the image, but overall the two images are very similar.

Comments welcome.

Rob.
---
Rob Fitzel
fitzel.ca/dart

Post a reply to this message

Attachments:
Download 'cabal4..jpg' (117 KB)

Preview of image 'cabal4..jpg'

"RobF" <nomail@nomail> wrote:
> This render took 6.8GB, mostly from the high resolution textures used on the
> characters. I experimented with both v3.6 and v3.7. The render took about 7h50m
> on v3.6. Surprisingly, the render took about 40 minutes with v3.7! On my Quad
> Core, I was only expecting a 4X increase, but it looks like there are some
> other cool optimizations in v3.7. There are a few radiosity artifacts in the
> 3.7 version of the image, but overall the two images are very similar.

I have found that it makes quite a difference whether I use the POV 3.6 Linux
binary, or the MegaPOV 1.2.1 Linux AMD64 binary - the is faster, despite having
the slightly more complex source code, so 64-bit optimized code gives some speed
advantage.

I also note that it makes a difference whether I compile the 3.7 beta with GNU
g++, or the Intel C++ compiler - again, the latter being faster.

So if you happen to be running the official Linux 3.6 binary, which is not
optimized for 64 bit and is compiled using g++, and the official Linux 3.7 beta
64-bit binary, which is 64-bit optimized and compiled with Intel C++, this may
account for quite some deal of speedup.

Post a reply to this message

"clipka" <nomail@nomail> wrote:

> So if you happen to be running the official Linux 3.6 binary, which is not
> optimized for 64 bit and is compiled using g++, and the official Linux 3.7 beta
> 64-bit binary, which is 64-bit optimized and compiled with Intel C++, this may
> account for quite some deal of speedup.

I used the Windows binaries from the povray.org website.

Rob.

Post a reply to this message

Impressive!

Post a reply to this message

"clipka" <nomail@nomail> wrote:
> "RobF" <nomail@nomail> wrote:
> > This render took 6.8GB, mostly from the high resolution textures used on the
> > characters. I experimented with both v3.6 and v3.7. The render took about 7h50m
> > on v3.6. Surprisingly, the render took about 40 minutes with v3.7! On my Quad

> So if you happen to be running the official Linux 3.6 binary, which is not
> optimized for 64 bit and is compiled using g++, and the official Linux 3.7 beta
> 64-bit binary, which is 64-bit optimized and compiled with Intel C++, this may
> account for quite some deal of speedup.

So, g++ gives 7h50m and icc gives 40 mins?  Those damned freetards! ;)

Post a reply to this message

"nemesis" <nam### [at] gmailcom> wrote:
> So, g++ gives 7h50m and icc gives 40 mins?  Those damned freetards! ;)

Naaah - it's not *that* bad. You forgot the QuadCore speedup.

Post a reply to this message

clipka wrote:
> "nemesis" <nam### [at] gmailcom> wrote:
>> So, g++ gives 7h50m and icc gives 40 mins?  Those damned freetards! ;)
> 
> Naaah - it's not *that* bad. You forgot the QuadCore speedup.
> 
> 

It's not only the multithreading and compiler. I have only one core (AMD
64 3500+) and I compiled my 3.6 from source, with gcc and probalby -O2
and -march=athlon64 (or w/e they are supposed to be). I think the
3.7-b29 was downloaded from povray.org. So, actually 3.6 should probably
be faster as it was optimized for this CPU, but no - 3.7 is (at least
for the "Eva Sails Away"-image) like ten times faster.

-- Arttu Voutilainen

Post a reply to this message

Arttu Voutilainen <blizzara.REM0VE7H!S### [at] zbxtSP4MM3Rnet> wrote:
> It's not only the multithreading and compiler. I have only one core (AMD
> 64 3500+) and I compiled my 3.6 from source, with gcc and probalby -O2
> and -march=athlon64 (or w/e they are supposed to be). I think the
> 3.7-b29 was downloaded from povray.org. So, actually 3.6 should probably
> be faster as it was optimized for this CPU, but no - 3.7 is (at least
> for the "Eva Sails Away"-image) like ten times faster.


Optimization for the official 3.7 beta 29 binary is probably -O3, and it comes
optimized for 64 bit as well.

"ten times faster" sounds a bit unlikely to me however. How much time are we
talking about?

Post a reply to this message

clipka wrote:
> 
> Optimization for the official 3.7 beta 29 binary is probably -O3, and it comes
> optimized for 64 bit as well.
> 
> "ten times faster" sounds a bit unlikely to me however. How much time are we
> talking about?
> 

I don't remember the times, but the difference was incredible. I guess
Arttu has them written down and will provide them, but I'll add
something: Eva sails away was also rendered (at least partly) on A64 X2
4450e (2,3GHz) with enough memory and C/CXXFLAGS "-march=athlon64 -O1
-pipe" with 3.6 and it took multiple times against the mentioned A64
3500+/3.7.

I also have one possible explanation: IIRC the image uses radiosity and
AFAIK the radiosity implementations differ between 3.6 and 3.7, being
able to create huge differences between render times. Feel free to
correct me if I'm wrong.

-Aero

Post a reply to this message

clipka wrote:
> Arttu Voutilainen <blizzara.REM0VE7H!S### [at] zbxtSP4MM3Rnet> wrote:
>> It's not only the multithreading and compiler. I have only one core (AMD
>> 64 3500+) and I compiled my 3.6 from source, with gcc and probalby -O2
>> and -march=athlon64 (or w/e they are supposed to be). I think the
>> 3.7-b29 was downloaded from povray.org. So, actually 3.6 should probably
>> be faster as it was optimized for this CPU, but no - 3.7 is (at least
>> for the "Eva Sails Away"-image) like ten times faster.
> 
> 
> Optimization for the official 3.7 beta 29 binary is probably -O3, and it comes
> optimized for 64 bit as well.
> 
> "ten times faster" sounds a bit unlikely to me however. How much time are we
> talking about?
> 
> 

I just tried: 3.6: 1min 27s vs 3.7b29: 5s

The images look different, but that is caused by different handling of
gamma. Otherwise I can't see any differences in them.

Both used the same scene and same ini. Radiosity was not used, nor focal
blur, nor media. I guess the difference comes from bounding boxes or
something like that, as IIRC it is caused mostly by grass.

-- Arttu Voutilainen

Post a reply to this message