Page 1 of 2

Comparison: Blender Internal Renderer, win32 vs. gentoo64

Posted: Thu Sep 08, 2005 11:33 am
by konrad_ha
As a short follow-up on my recent Yafray win32 vs. gentoo64 comparison I now had the chance to compare rendering-times with the Blender's internal renderer:

my short specs:
AMD Athlon 64 3400+
1 GB Ram

The scene contains some very large textures and is lit by a lightdome casting buffered shadows. No raytracing is involved. The OSA is set to 16. Here's the output, the rendering times follow below:
Image

the times:

- win32: 1min 27sec
- gentoo64: 48sec

Before the usual "unfair comparison"-posting starts let me mention my intention (ha!):

This test is intersting as it compares stable, tested and non-customized systems and software-builds. Everything I used works out of the box (if you consider gentoo "out of the box") and is proven to be stable and secure within a production-environment.

For people like me, who prefer to use software rather then compile it this comparison might be meaningful.

Cheers,

Konrad

Posted: Thu Sep 08, 2005 12:43 pm
by Dolphin
Thanks for the comparison, it has really interessting results.

But I miss some information that I think is important to interpret the results correctly:

1. Which blender versions did you compare?
2. Which gcc version did you use with gentoo?
3. Which compiler options were set for gcc?
4. How much free RAM was available before you started rendering?
5. Did both operating systems ran in 64bit mode?

I assume you used the installation package from blender.org for the Windows version and compiled blender from source with gentoo. So I think it is no suprise that the gentoo version is more optimized.

Therefore it would be nice to see an extended comparison that includes:
1. Optimized Windows version (Intel compiler or same gcc version?)
2. Standard Linux RPM package for blender

Best regards,

Dolphin

Posted: Thu Sep 08, 2005 2:39 pm
by joeri
So, are both images the same?
Can we see a black diff image?

Was the rendering done in the background (don't want no "put image to screen" bias :) ) ?

And, yes, did you surf the web while rendering or where the machines running only blender and left alone? Or is multitasking a part of the test?

Forget about the gcc comments, i thought you made it pretty clear it's not something downloaders want to worry about. True?

Posted: Thu Sep 08, 2005 3:46 pm
by wavk
Another important thing is did you render from command line or from the GUI? Different video drivers/os's can have different drawing speeds. Drawing rendered pixels in the render window might make a big difference.

Posted: Thu Sep 08, 2005 3:50 pm
by konrad_ha
Here's the info I can provide:
1. Which blender versions did you compare?
2.37a on both systems
2. Which gcc version did you use with gentoo?
The one that comes with it?
3. Which compiler options were set for gcc?
The ones recommended in the gentoo installation docs I think.
4. How much free RAM was available before you started rendering?
While I can't tell you exactly how much ram was free I can assure you only the absolute minimal necessary background-tasks were running on both OSs during rendering. This also answer the question whether the machines did anything else while rendering: of course not.
5. Did both operating systems ran in 64bit mode?
No, win32 did not run in 64bitmode, duh!
So, are both images the same?
Can we see a black diff image?
Yup, I immediately compared the images with Gimp after rendering, overlaying them with the difference-mode, and the result was ... black. There's just no difference at all.

And again I need to point out: I'm pretty sure one could build a faster Blender-version on win64, but as joeri says: it's not something I feel capable of doing. I'd rather enjoy twice the speed I get with a simple "emerge blender" command on gentoo.

Posted: Thu Sep 08, 2005 7:29 pm
by joeri
Cool.
So, was it in the background (commandline) or Gui ?
I noticed blender got a bit slower (on windows) after the display window got more advanced.
I heard slid3 saying commandline is up to 30% faster than the gui, I have no idea if this is (still) true or not.

Posted: Fri Sep 09, 2005 9:31 am
by Dolphin
Thanks for your replies konrad_ha and sorry for the stupid questions I asked :oops: (especially the win32 / win64 one...)

Posted: Fri Sep 09, 2005 4:22 pm
by TheFallenWeeble
konrad_ha wrote:
2. Which gcc version did you use with gentoo?
The one that comes with it?
Typing "gcc --version" should give you a more definitive answer.

Posted: Fri Sep 09, 2005 4:34 pm
by joeri
TheFallenWeeble wrote:
konrad_ha wrote:
2. Which gcc version did you use with gentoo?
The one that comes with it?
Typing "gcc --version" should give you a more definitive answer.
Really? Where, in notepad? Or do you think windows comes with gcc installed?

Posted: Fri Sep 09, 2005 4:37 pm
by kidb
joeri wrote:
TheFallenWeeble wrote:
konrad_ha wrote: The one that comes with it?
Typing "gcc --version" should give you a more definitive answer.
Really? Where, in notepad? Or do you think windows comes with gcc installed?
tststs

Posted: Fri Sep 09, 2005 4:41 pm
by joeri
To harsh? I apolagize for that.
I thought it was clear it's a 'consumers' (out of the box) test.
blender consumers don't install gcc.

Posted: Fri Sep 09, 2005 7:24 pm
by fearandloathing
joeri wrote:
TheFallenWeeble wrote:
konrad_ha wrote: The one that comes with it?
Typing "gcc --version" should give you a more definitive answer.
Really? Where, in notepad? Or do you think windows comes with gcc installed?
Since when was there a gentoo windows distribution? He asked wich gcc version he used with gentoo, not windows.
joeri wrote:To harsh? I apolagize for that.
I thought it was clear it's a 'consumers' (out of the box) test.
blender consumers don't install gcc.
We're talking gentoo GNU/Linux here.
Perhaps you could live without gcc with some binary package distributions like fedora, but not gentoo (sure you can probably install binary packages for many things, but the nature of this distro is to compile your own packages).
Even if it's a standard gentoo install with the default compiler options, he could still answer that question by typing gcc --version in the console.

Posted: Fri Sep 09, 2005 7:44 pm
by joeri
My mistake.
Makes the 'only download' part of the test a bit silly, when running on a 'only compile' os....
I mean, why claim it's an "out of the box" test if you do a compile for 1 os?

Posted: Fri Sep 09, 2005 8:28 pm
by fearandloathing
joeri wrote:My mistake.
Makes the 'only download' part of the test a bit silly, when running on a 'only compile' os....
I mean, why claim it's an "out of the box" test if you do a compile for 1 os?
It's the nature of the distribution, you compile your own packages.
If you want to install blender on gentoo you simply issue the command to automatically download the source package, compile and install it (I think it's emerge [package] on gentoo, never used the distribution myself).

To be fair though, he should have compared it to Windows XP x64, comparing 64bit gentoo to 32bit windows on a Athlon64 isn't fair (not sure about the performance difference here).

Although most users coming from windows will probably prefer a binary package distribution like fedora or ubuntu, using such a distro for a benchmark comparison would produce results closer to windows.
But I guess most people choose gentoo because of the performance increase resulting from compiling everything from source optimized for their own hardware.

Posted: Sat Sep 10, 2005 1:05 am
by joeri
Well I didn't know what gentoo was and find the

"For people like me, who prefer to use software rather then compile it this comparison might be meaningful. "

misleading. As turns out that 1 of the test runs was custom compiled.