I always knew it: Blender + Yafray MUCH faster on Linux!

General discussion about the development of the open source Blender

Moderators: jesterKing, stiv

konrad_ha
Posts: 0
Joined: Wed Jul 21, 2004 12:15 am
Location: Munich | Germany | Europe
Contact:

I always knew it: Blender + Yafray MUCH faster on Linux!

Post by konrad_ha » Wed Aug 31, 2005 1:04 pm

Hi there,

I have to start rendering my latest project the coming days. So I figured it was finally time to start comparing Windows to Linux as rendering-environment (after having gone through all the trouble of learning Gentoo, of all distros!).

I'll make more thorough comparisons later, but the first test hit me like a hammer.

The scene contains transparency, image- and procedural textures, an omni light for highlights and is lit by the Yafray Skydome method. Object complexitiy is pretty low. Now the results:

system 1
AMD Athlon 64 3000+
1GB Ram
Gentoo Linux 64
Kernel 2.6.12
time: 02:04.11

system 2
AMD Athlon 64 3400+
1GB Ram
Windows XP SP2 (not 64bit)
time: 02:54.67

So rendering on a faster Processor with 32bit Windows is about 30% slower than on a highly optimized 64bit Linux architecture on an actually slower system.

This was just the first shot, I'll soon compare rendering times on the same machine (system 2) where I have a dual-boot installation.

But even know I am suprised at how big the performance-gap is. It's just incredible!

Do you have other data for performance comparison? I'd really like to study this whole issue with more depht, probably even write a short article for the Digital Production about it.

Render on,

Konrad

nemyax
Posts: 0
Joined: Mon May 31, 2004 12:34 pm

Post by nemyax » Wed Aug 31, 2005 1:49 pm

Only goes to show that a 64-bit OS is faster than a 32-bit one. There's nothing more to it.

konrad_ha
Posts: 0
Joined: Wed Jul 21, 2004 12:15 am
Location: Munich | Germany | Europe
Contact:

Post by konrad_ha » Wed Aug 31, 2005 8:37 pm

I think it is indeed very interesting.

I just happened to recently read an article comparing Win32 and Win64 performance. It showed performance-gains between 1% and 10% for floating-point operations. The peak performance increase was visible during media-encoding (15.5%) with XMpeg. Though still in Beta-Stage I doubt Win64-final will surprise us with big performance-gains in any particular field.

When I compare this to a performance-gain of almost 30% during raytracing on gentoo64 I can't help but being astonished.

Searching turned up another report regarding x86_64 performance on Windows. It's an interesting read especially as POV-Ray is included in the testing, the 32bit as well as native the 64bit version. Again the performacne-gain of the native 64bit version over the 32bit version (both tested on an AMD64) isn't even near the 30% I could witness on my systems.

I'm already looking forward to testing rendering-times on the same machine, where I can get some clearer results.

I'd also be eager to get some comparable rendering-times with this scene on Win64, but I'm not even sure whether Blender and Yafray would even run on it.

qwe
Posts: 0
Joined: Wed Jan 19, 2005 7:35 am

Post by qwe » Wed Aug 31, 2005 11:33 pm

this comparison is absolutely worthless. You can't compare applications running in 32-bit mode on a 64-bit processor with one OS to another OS running in 64-bit native mode and conclude that the 64-bit native OS is "faster". All you've done is prove that 64-bit native applications run faster than 32-bit applications on 64-bit processors.

dumbf
Posts: 0
Joined: Wed Aug 31, 2005 2:16 pm

Post by dumbf » Thu Sep 01, 2005 12:00 am

ok I might be jumping to conclusions here but it sounds like there's some anti linux users here.

i can't see that what konrad_ha is doing is worthless. what it shows is if you've got a 64bit processor and you want to create a render node or nodes to run blender+yafray blender+pixie blender+aquis as cheap as possible then linux 64bit and you render soft compiled for 64bit is free against forking out for WinXP 64bit if your on XP 32bit already. on top of that performance gain for free you've also got the other benefits of linux for render nodes, why else do cg companies build render farms on linux?

qwe
Posts: 0
Joined: Wed Jan 19, 2005 7:35 am

Post by qwe » Thu Sep 01, 2005 12:32 am

dumbf wrote:ok I might be jumping to conclusions here but it sounds like there's some anti linux users here.
you think I'm "anti-linux"? Let me inform you that I run Gentoo 100% of the time. I don't even have Windows installed on my computer. And you're right, running a linux-based renderfarm is extremely cost-effective.

However, cost-effectiveness was not his point. He compared performance between 64-bit linux and 32-bit windows on the same processor. It's like comparing apples and oranges. If you want to do some benchmarking, either compare 64-bit linux and 64-bit windows or 32-bit linux and 32-bit windows.

dumbf
Posts: 0
Joined: Wed Aug 31, 2005 2:16 pm

Post by dumbf » Thu Sep 01, 2005 1:04 am

hey qwe that's great 100% Gentoo Linux.

My mistake was that I didn't see that what Konrad was doing was benchmarking, i saw it as drawing comparisons, he didn't mention benchmarking?.

Like: I have a 64bit machine with Xp on it and I have a 64bit machine which I could install a 64bit Linux on it and try for nothing. Now which one is going to give me the best performance with the apps I want to run (blender, yafray etc) and the purpose (render node) with what I've got. It's easy to think a 64bit machine is going to be faster than a 32bit but that's not necessarily true or not a big enough perfomance gain to make it worthwhile for the outlay.

i think worthless is too harsh for what he appears to be doing.

cekuhnen
Posts: 0
Joined: Mon Jan 13, 2003 11:04 pm

Post by cekuhnen » Thu Sep 01, 2005 1:18 am

well

as far as i know and could see it 64bit does not render anything faster.
the hole bit theory is based on accessing the ram. well when you run
out of ram or the image is to big to handle i guess sure 64bit might
seem faster, but thats not the integer unit that makes the speed
increasement, it is basicly the memory access unit.

i guess even with the 64k windows version the speed will not be as much
faster as long as the scene does not exhaust the memory.

anyway his linux pc is also slower.

qwe
Posts: 0
Joined: Wed Jan 19, 2005 7:35 am

Post by qwe » Thu Sep 01, 2005 2:59 am

You people are missing the point. Konrad_ha is trying to use this comparison to prove that Linux is somehow inherently faster than Windows since 64-bit Linux is faster than 32-bit windows on a 64-bit processor, which, as I said, is like comparing apples and oranges.

I don't deny that Linux is faster than windows in most cases, but the evidence does not support the hypothesis.

cekuhnen
Posts: 0
Joined: Mon Jan 13, 2003 11:04 pm

Post by cekuhnen » Thu Sep 01, 2005 3:32 am

he also mentioned that the complexity is pretty low, so i guess that
he will not cross that border in which the 64 bit access to more ram
will have an advantage compared to the 32 bit access to ram.

however your rejection can be somewhat right ;)

LetterRip
Posts: 0
Joined: Thu Mar 25, 2004 7:03 am

Post by LetterRip » Thu Sep 01, 2005 1:30 pm

I seem to recall that AMD64 has additional registers that give it a significant performance boost for things like rendering, as others have pointed out to make it a fair comparison, you'll need to compare performance using the OSes on the same hardware.

LetterRip

joeri
Posts: 96
Joined: Fri Jan 10, 2003 6:41 pm
Contact:

Post by joeri » Thu Sep 01, 2005 1:39 pm

He now, I would like to know what happens on the same machine.

I don't mind the comparison of oranges with apples if the tells me that oranges render faster (for the same price or cheaper).

So, same machine, different OS's and then tell me which OS is fastest.
Then; other machine, different OS's and then tell me which OS is fastest on that machine.

and then post it on the yafray forum, 'cause yafray render speeds is not really something blender can do anything about.

winfried
Posts: 0
Joined: Mon Dec 06, 2004 2:56 pm

Post by winfried » Thu Sep 01, 2005 1:54 pm

One thing to remember: if you compile with gcc on a 64 bit platform, gcc generates sse(2) instructions by default. On a 32 bits platform you have to enable this manually (and yafray build doesn't). So to make a comparison that's more fair, you should recompile yafray on windows with sse enabled.

But we all know linux is faster :wink:

joeri
Posts: 96
Joined: Fri Jan 10, 2003 6:41 pm
Contact:

Post by joeri » Thu Sep 01, 2005 2:23 pm

gcc?
Is that the linux compiler that got ported to windows?

konrad_ha
Posts: 0
Joined: Wed Jul 21, 2004 12:15 am
Location: Munich | Germany | Europe
Contact:

Post by konrad_ha » Fri Sep 02, 2005 4:33 pm

...and now to the direct comparison:

system:
AMD Athlon 64 3400+
1GB Ram

A new scene showed the following results (rendering times for a single frame):

Windows XP 32 SP-2: 4:05.49
gentoo 64 kernel 2.6.11: 2.42.49

I leave it to you whether you consider these numbers "absolutely worthless".

As far as I am concerned it highlights the difference between two standard and proven systems on the same hardware. It's also worth mentioning that Yafray was aquired via standard ways of distribution and I didn't have to go through the hassle of compiling it myself (portage does that for me).

It may be possible to reach the same high performance with a Yafray SSE2-64bit build on Win64, but I would neither want to run Win64 in it's current state nor would I want to compile Yafray myself (especially not on Windows).

Post Reply