Benchmark rendertimes

[gunnarstahl]

Hi,
I thought about doing some community made benchmarkings. This came to my mind last week when reading the particle grass thread in finished projects.

I took the grass_tutorial blend file (http://www.cibiv.univie.ac.at/%7Egre...rial.blend.zip) and it took about 4:15 minutes just to render the first frame of this animation on my iBook 1.2GHz. I wondered how long it would take to render it on a decent machine.
So I took it to my machine at work (3.06 GHz Hyperthreading, 2 GB ram) to render it.

To my total astonishment it took exactly the same time to render it. 4:15 minutes! When turning on threads for rendering the rendertime dropped to 3:30 minutes.

So I would like to hear from you people how long the first frame of the mentioned file takes to render on your machines!

To have similar circumstances please stick to the following procedure: Open the file in blender and immediately press F12 to render it. Then enable "Threads" in the rendersection and press F12 again to render it.

Post the results here with the specs of your machine.

Thanks,
Gunnar

[Mike_S]

Originally Posted by gunnarstahl Hi, I took the grass_tutorial blend file (http://www.cibiv.univie.ac.at/%7Egre...rial.blend.zip) and it took about 4:15 minutes just to render the first frame of this animation on my iBook 1.2GHz. I wondered how long it would take to render it on a decent machine. So I took it to my machine at work (3.06 GHz Hyperthreading, 2 GB ram) to render it. To my total astonishment it took exactly the same time to render it. 4:15 minutes! When turning on threads for rendering the rendertime dropped to 3:30 minutes.

How much ram do you have in your IBook ?

According to Broken (Matt Ebb of Project Orange), he has said that ram is more influential on rendering times than cpu speed. Though with 2g on your work machine (and I'm guessing ??? less) on your Ibook, the results are interesting.

I'll try it on my meagre 1.2g 256meg machine and let you know.

Do you have any of the Elephants Dream files? Try rendering one of those on your work machine


Mike

*Edit* 8:55:50 to render ...but I have a *lot* of other stuff running right now at the same time ... I'll try it again later with nothing else running, but I'll guess it won't be any faster than your 4:15 time.

[gunnarstahl]

The iBook has 768 MB of ram. So even this should make it a much less decent numbercruncher...

[RJii]

On an Amd2000+, 1Gig, WinXpSp2 machine with a few things in the background the original image was 3:56. With threads 4:19 (which was ofcourse to be expected as the Athlon doesn't have any HT or multiple cores).

[chop_suey]

AMD Athlon 2400+, 1Gig Ram, WinXpSp2 with foobar2000 in the background:
exactly 2:31
with threads: 2:30 (strange?!)

edit: maybe the dual channel mode might affect the speed? 2x512mb RAMs

[Lynx3d]

Wow i must have a speed demon then...and i already wanted to replace my machine

non-threaded: 2:07
threaded: 1:38 (though with more than 2x2 parts it could be a bit faster)

somewhat fresh cvs bf-blender, linux (ubuntu 6.06), dual-Athlon MP 1800+, 1GB RAM.
But i assume, it's once more the linux-factor...at least with yafray linux binaries beat windows binaries so badly that i keep wondering what we are doing wrong...

[DadCantDraw]

AMD Sempron 2000MHz 512MB RAM Windows XP SP2 Blender 2.41 and running IE in the background:

Non thread: 2:15:20
Thread : 2:14:59

<edit> I just recently ran services.msc to get rid of a bunch of junk Winblows runs by default <grin>

[Mike_S]

Originally Posted by Lynx3d ..at least with yafray linux binaries beat windows binaries so badly that i keep wondering what we are doing wrong...

If "what we are doing wrong", you mean, "why aren't more people using Linux instead of windows", the answer is simply that Windoze remains much more easier to setup on average hardware, and so are the applications. I installed Linux about 7 months ago, and it was a recent release at the time. It still didn't have a driver for my very generic Geforce video card, I had to go 'net searching for it. Finding DVD player software was also a chore .. and the list goes on and on.

A highly tweaked Linux box may blow the doors off a "tweaked Windoze box (if there is such a thing), but for the hours of work it takes, the average user will just accept the lower performance.

Mike

[gunnarstahl]

Does anybody have one of these new shiny dualcore macs? Would be nice to know how these "4-5 times faster" beasts really perform...

Gunnar

[Lynx3d]

Originally Posted by mstram If "what we are doing wrong", you mean, "why aren't more people using Linux instead of windows"

No i meant "we as developer" (of yafray)...
The OS should not make a difference of 30% or more in performance, but the difference is there on my machine. I already tried to do as much manual memory managment as possible etc. but the gap is still large. It's also not GCC vs. MSVC because MinGW produces just as slow binaries.

[Le_Mackeux]

I don't have my MacBook Pro yet but, I ran it to my dual G4.
The machine is a DualProcessor 1Ghz G4 with 1Gb pc133 ram.


Doing the test without any changes, it took 3:58
Doing it with the threads option, it took 3:22


I'll come back here to post my results with the MBPro when I'll have it. About 2 weeks.

[tedi]

http://www.eofw.org/bench/

[etr9j]

Here is my first machine - Sorry for all the times, I get carried away with things like this:

Blender 2.41, Blender CVS 6/11/06, 2.41 SSE2 Optimized, (couldn't get Blender x64 to load the file)
Pentium D 930, dual core, 3.0 GHz, 800MHz FSB, 2x2M L2
2 Gigs, DDR2 @ 533 MHz, Dual Channel
Windows x64

Render Time without threads
2.41: 2:10.58
CVS: 2:07.87
2.41 w/ SSE2 optimization: 1:08.76
2.41 w/ SSE2 optimization xparts=1 and yparts=1: 1:08.33

Render Time with threads
Blender 2.41: 1:52.37
Blender CVS: 1:36.41
2.41 w/ SSE2 optimization: 0:51.04 - sweet
2.41 w/ SSE2 optimization xparts=1 and yparts=1: 1:08.04

As for xparts=1 and yparts=1, it's actually slower or it stays about the same.

Conclusion? For Windows builds, optimized is a must!

[dschnell289]

Rev A iMac G5 1.8 GHz, 1 GB DDR400 RAM, OS X 10.3.9, Py 2.35

2:05.74 as is Blender 2.41
1:57.91 as is but with xparts=1 and yparts=1 - usually is faster. (threaded version of this is 1:59, as expected).

2:28.53 as is with April CVS version - can't reduce xparts and yparts below 2.

[broken]

Originally Posted by mstram According to Broken (Matt Ebb of Project Orange), he has said that ram is more influential on rendering times than cpu speed. Though with 2g on your work machine (and I'm guessing ??? less) on your Ibook, the results are interesting.

That was with regards to the Elephants Dream files, since we used a lot of very large texture maps and shadow buffers (which take lots of RAM). Other scenes may be more reliant on CPU, it really depends on the scene itself. Following that, a good benchmark would contain a variety of different scenes focused on different areas, to test different parts of the system. Large texture maps or video textures, lots of procedural textures, high raytracing depth, dense geometry, complex animation, large scales, all can be influenced by different parts of the system.

[Lynx3d]

/me smiles that the Pentium-D 930 with twice the clock frequency and 4 times more cache is a whole second faster...
Okay i cheated, i changed my compiler flags to "-O3 -march=athlon-xp"...


And, unless memory usage exceeds free RAM, CPU is pretty much all that matters (well memory bandwidth and latency a bit too), otherwise your HD and virtual memory sub-system is all that matters...because your CPU will get bored at a few percent load waiting for the constant paging...
Well cases between those extremes exist too of course, e.g. when only part of the whole rendering process needs more memory than you have free RAM, or a large enough amount of the generated data is not accessed for long times.

[etr9j]

Originally Posted by Lynx3d /me smiles that the Pentium-D 930 with twice the clock frequency and 4 times more cache is a whole second faster...

/me

It is my fond hope that maybe my times were a statisitcal outlier, but I don't think that was the case. Maybe it's the Linux (as mentioned above) or the AMD, but either way I am severely disappointed in my machine's performance! And given the time I see below for my work laptop, I am thinking I need to test using Linux on the same machine.

Here is my work laptop:

Blender CVS 6/11/06
Pentium M, single core (no HT), 1.8 GHz
1 Gig DDR
Windows XP Pro

Render Time without threads
Blender CVS: 2:13.13

Render Time with threads
Blender CVS: 2:12.98

*EDIT* 1:11.76 with 2.41 SSE2 Optimized Windows (graphicall) version!

[eeshlo]

Strange, sofar it seems I beat you all, and my system is really nothing special, tested with cvs build:
without threads: 1:09:93 and with threads: 1:09:85, of course, no difference there because I only have one processor:
Athlon 3500+ (2.2G), only 512MB ram (DDR something..., not sure what), SuSe Linux version 10.0, 64-bit though, so maybe that explains the difference, but I only used the default compile settings.

[Fliberty]

Well, my no threads is a little slower since my cpus are only clocked at 2.1 GHZ, but my threads kicks arse.

No Threads:

1:11.88

Threads:

0:59.37

Dual Opteron 270 Dual Core (4 Cores Total 2.1 GHZ each) 6 X 512 MB Dual Channel DDR RAM ( 3GB )

64 bit Gentoo Linux

Oh yeah, for more E-Peen 4 x 250 GB 7200 RPM HDDs coming to a whopping 1 TB. Still havent found enough stuff to fill it. ;( Haha, I'm so mean.

Not as good as i had hoped. Seems like its only using half of my cpu when i render. So, I'll have to look into that.

[p00f]

Dual opteron single core 1.4 Ghz overclocked to 1.6 Ghz. 2 gigs 333hmz DDR per processor (dual bank) tyan mobo.

all times are using the SMP threading...
Running Windows X64
blender 2.41 32 bit 2:47.35
blender 2.41 64bit 2:40.21
blender 2.42 32 bit bullet game engine preview 2:47.34

Running Kanotix64 (Debian with super cow powers!)
blender 2.41 from apt repository 2:00.12

Running Ubuntu64
blender 2.41 from apt repository 3:30.32