blender is better in windows 7

Support Compositing and Post Processing
1 
			   						 							 							 						 						 							 						 				 					 						 							I've installed blender in a computer with two operating systems  working on it: ubuntu 10.10 and windows. I've get a model from here:

nextwavemultimedia.com/html/3dblendermodel, the “sofaset” is the one I chose.

I’ve made a test of rendering time of that model in the last released beta version of blender. In ubuntu it took a rendering time of 27 minutes, 5 seconds. In windows it took 14 minutes with 45 seconds.
in version 2.49 and it was worst. it isn’t supposed to be faster blender in linux? I didn’t change anything of the software’s configuration. Someone can tell me why this happened and how to get a better performance of blender in linux?

thanks.

2

Before doing anythng, get your graphics drivers up date and make sure hardware acceleration is working properly in Ubuntu.

To get Blender running the fastest on Ubuntu, I build it myself from svn. There are instructions on Blender.org that tell you how to build using 2 different systems. I use scons because it is included as a part of Blender’s source code and it’s super easy to automate. You can find instructions here: http://wiki.blender.org/index.php/Dev:2.5/Doc/Building_Blender/Linux
Before compiling, you will need to make your own user-config.py file as mentioned in the above tutorial. Here is my user-config.py contents, as an example:



BF_BUILDDIR = r'../buildOpt' 
BF_INSTALLDIR = r'../installOpt' 
WITH_BF_OPENAL = True 
WITH_BF_QUICKTIME = False 
WITH_BF_GAMEENGINE = True 
WITH_BF_PLAYER = False # set this to True whenever the player has been ported properly 
WITH_BF_JACK = True 
WITH_BF_SDL = True 
WITH_BF_ICONV = False # has something to do with Collada says jeff 
WITH_BF_COLLADA = False 
WITH_BF_OPENEXR = True 
WITH_BF_FFTW3 = True 
WITH_BUILDINFO = True 
WITH_BF_RAYOPTIMIZATION = True 
WITH_OPENMP = True 
WITH_BF_TIFF = True 
BF_NUMJOBS = 2 
 
REL_CFLAGS = ['-O3','-fomit-frame-pointer','-funroll-loops','-ffast-math' ] 
REL_CCFLAGS = ['-O3','-fomit-frame-pointer','-funroll-loops','-ffast-math' ] 
 
CCFLAGS = ['-pipe','-march=nocona','-ftracer','-ffast-math','-msse','-msse2','-msse3','-mmmx','-funroll-loops','-fforce-addr','-fomit-frame-pointer','-funsigned-char','-fno-strict-aliasing' ] 
CXXFLAGS = ['-pipe','-march=nocona','-ftracer','-ffast-math','-msse','-msse2','-msse3','-mmmx','-funroll-loops','-fforce-addr','-fomit-frame-pointer','-funsigned-char','-fno-strict-aliasing' ]

You will need to research GCC optimisation switches in order to get the best results for your CPU. The knowledge you learn here will allow you to build many applications yourself and to see a performance boost in the applications you’ve optimized. I must admit that I have only gone through the trouble of doing this with Blender though!

You can also try the builds on Graphicall.org. Look for a build in the right column that is optimized for your CPU.

3

Really in my experiments I found little to no difference between the Linux and Windows. People will proclaim one or the other as being better, but realistically if you build Blender with optimizations specifically for your system results will be nearly identical.

4

BF_NUMJOBS = 2

Of course, set this number depending on the number of cpu cores that you have.

5

There is a great difference if you compile the code with some options than with other options (for example -02 for optimization) and also there is a difference with the compiler you uses. And also if the computer is AMD or Intel. But in this extreme case of near double time I think you are testing two very different builds, one probably without optimizations and the other with them. Perhaps even you are using a non official build versus an official build.

I would like to test myself, could you say where I can find the blender scene? I tried: http://www.nextwavemultimedia.com/html/3dblendermodel
and that page don’t exists.

6

Maybe some other program? there will be huge difference in performance if for example you have heavy flash in one of firefox tabs in background.

And in ubuntu there are some awful IO bugs which affects performance. Because of this I use optimized gentoo as my OS for rendering and drawing (and other heavy computing tasks).

No Windows version ever can beat custom build Linux kernel specially for rendering and with ultra light weight GUI for desktop. And of course no background processes!

7

I found it here:

(www) nextwavemultimedia.com/html/3dblendermodel.html

Sofa Set is about halfway down. I’ve been meaning to test it myself…but man I hate booting into Windows :wink:

Let me know your results.

8

I was testing myself with a realistic head and:
-Ubuntu64: Blender 2.56a: 2minutes18seconds
-Windows7 64bits: Blender 2.56a: 2minutes27seconds

So I find it quicker yet in ubuntu. Also I find the mouse and handling meshes with million of polygons better in ubuntu. I have two hard disks with ubuntu in one and windows7 in the other so the same hardware.

A very important note in my opinion for the 2.56a blender beta for 64bits:
In windows you can see the blender executable that has a size of 24Mb. And a lot of .dll in the same directory. (For non programmers, dll are a way of store external programs that blender calls when he needs things like sound, open image formats,…).
In linux the blender executable has a size of 50Mb and you don’t see any .so (that is the extension the .dll have in linux) so all these libraries are inside the blender executable. This is not good thing. When the code grows you spend more time just to run it. I build blender in ubuntu and my blender executable has a size of 21Mb. All the .so are not in blender executable but out of it. The render time with my build was of 2minutes7seconds. 11 seconds less than official build. I prefer my build.

Thanks for the link to the scene. I will test and post here my result with the sofa scene.

9

Just tried the sofa scene (open the scene and F12 without changing nothing on it):
Ubuntu 64bits: 1minute50seconds
Windows7 64bits: 2minutes4seconds

Another thing: When you enters ubuntu, the memory used before opening any app, just the memory ubuntu needs to run is 500Mb
Windows needs 1.2Gb just to run. So you always have 700Mb more of memory for your programs in ubuntu. This is a very important point for people that only has 4Gb of memory or less.

So my guess is that the creator of this thread is using ubuntu 32bits and comparing it to windows7 64bits. Or he is running a graphicall build not the official build. But I can see ubuntu is better yet.

10

I think I know what happened!
If you compile in ubuntu 64 with scons you get a blender much quicker than if you compile with cmake.
I just remembered this, I was testing some weeks ago because Campbell Barton posted about how to build blender in QtCreator and cmake. Well, render times were almost double so I continue building blender with scons. It must be some compile options that are different in both methods but I couldn’t find it.
So my guess is that the poster of this thread is using a graphicall build for ubuntu created with cmake. Use the official build instead (2.56a in blender.org) and tell us what render times are then.

11

hi!

i’m so sorry to answer until now. thanks for all your answers, but specially to loopduplicate. i´ve made now a compilation of blender in my computer and i get better results in linux. it is almost at the same speed than in windows and sometimes faster.

now i’m getting started with some scenes of myself. i’m making a cup modeleddd in 3ds max. i’ve exported it in stl and the goal is to obtain similar rendering results in blender; to get similar images. at that point i have some new issue: the stl imported onto blender is not very smooth. i’ve tried with the smooth modifier, but it disappears almost all the mesh, leaving only dots or something like a “wire netting” for a cup. but well, until now it isn’t so bad. my interest at first point is to prove rendering and slowly i’ll move into modelling.

again thanks, and i apologize myself to post in different forums the same doubts. in the future i’ll try to go specifically to the right place.