Windows VS Linux

I have a question about the performance differences in Blender 245 for Windows and Blender 245 for Linux.

Machine #1: P4 3.4GHz 800MHz FSB 2Gig RAM running 2 threads with WinXP
Machine #2: HP DL380 G3 server - 2 X xeon 2.8GHz 400MHz FSB 2Gig RAM running 4 threads with Red Hat Fedora 8

I assumed that I would get nearly equivalent performance but I did a test render of the exact same scene on each, I verified that both threads on the XP machine were in use during render, and that 4 threads were in use during render on the Fedora machine.

The XP machine rendered the scene in 13mins 04secs while the Fedora machine took 20mins 04secs.

That puts the 4 X 2.8GHz/400MHz threads at about 65% of the 2 X 3.4GHz/800MHz threads performance wise.

The scene was using Blender Internal.

If I installed WinXP on the server would I see an improvement in performance regarding Blender rendering?

Does the registered memory of the server make that large of an impact on the FSB?


What was the nice value used on the Fedora machine when rendering?
Any optimized builds used?
And was it rendered through the command line?

I found the biggest boost for linux is when switching to 64 bit. Linux in 64 bit means everything is true 64 bit, not just bits and pieces as per Windows.
Then in Linux, you can use optimised builds and raise the priority as Kit89 mentioned. To optimise fully you would need a meta-distro like Gentoo, and recompile everything. ( I didn’t find that worth my time. I switched back to Ubuntu. It just works.)

I changed nothing from default install of Fedora 8 using Gnome desktop, did yum update so all installed packages are current.

I installed Blender 245 using the package manager, so it is the build that comes in the standard repository.

I ran the render normally via opening Blender to the desktop.

Before starting render the 4 threads were approx. 1% load, during render all 4 threads were peak at 100% load.

I wanted to make the comparison as equal as I could so I did the render the same way on both machines with desktop GUI and nothing done beforehand to either machine to squeeze out more performance.

While I really had better hopes for the performance of this server, I picked up the complete DL380 G3 with both CPU’s installed and 6 X 18.2Gig 15K HDD’s and the 2Gig’s of RAM for $300 so I guess I won’t slash a wrist over it, and come to think of it I believe I can easily disable ECC on the RAM in the BIOS, ok, I’ll try that and report the difference.

EDIT: Nope, can’t disable ECC.

This might interest you…

Good link egan. I’ll may switch to Linux at some stage.

hehe, locked :smiley:

I’ve done multiple tests between Windows XP 32bit and Ubuntu 64bit.

Ubuntu won on all the renders by about 60%. So, over twice as fast.

EDITED: All OS’s are equal, and we are all crazy.

nope :frowning: i wish

Wow, so whats the difference then? I cant be making these incorect statements all over the place…

that sounds about right to me, but in reality thats not true about any OS users, generalizing wont get anybody anywhere… :frowning:

Hahaha, that was a very slick edit, I had to go back and check my own post…

64-bit relates to register size and data path width, not processor speed. 64-bits are somewhat faster in things dealing with high precision or lots of data processing (which is why they’ve been implemented in super computers for decades).

I was able to get the best rendering performance using Gentoo Linux in 64bit mode, everything custom compiled, with 4 gigs of memory.

That said, any serious bit of rendering really is best done on a render farm. It’s a hell of a lot cheaper than buying all the hardware you would need to do it at home.

I would just recommend using whatever OS you want to use.

Dats sum BIG generalizashums, buddy.
Windows users can be just as fanatical Just check out some of the trolls you get on linux message boards! Whew!
Same with Mac users. Some of those dudes are downright evangelical about it.
Anyone who gets fanatic about their OS oughta get a life. While I use Windows at work, I use Linux at home. I made my choice and I’m happy with it. You can use an abacus for all I care. (As long as you can compile Blender on it!)
It’s simply not worth getting passionate about.

Back to the topic, I definitely find I get better performance under Linux. So therefore it is THE BEST DO YOU HEAR ME?!! THE BEST!!! MWA HA HA HA HA HA!!!

The only true test is to run Blender on the same hardware with different OSs. I’ve tested Windows and OS X on the same hardware and they come out almost exactly the same after multiple tests. The optimized Windows build was faster as you’d expect.

Optimizing the Linux binary and switching to 64-bit is ridiculous if you are only trying to just match an unoptimized 32-bit Windows binary on a machine that should be about half the speed (dual 2.8 Xeon vs single 3.4 Pentium and the Xeon is half the speed?? It should be the other way round).

If Linux isn’t performing well then don’t use it, use what does perform well on the hardware you have.

At least now we know if the Peach team had rendered using Windows they’d have finished ages ago. :wink:

Hahahahaha, I guess I shouldnt be so general in my statements. Edited for politcal correctness…

While I am currently experimenting with the DL380 server I had thought about installing XP on it for a completely level comparison of hardware, but given that the server FSB is 400MHz and the RAM is ECC enabled, the server appears to have the expected trade-off of reliability over pure performance.

I thank you all for weighing in, got a couple of nice links out of it and some interesting info.