Benchmark Test Results Using Blender 2.5 Alpha 2
My name is Ivan Paulos Tomé (A.K.A. Greylica), I’m the Maintainer of the Brazilian Wiki and I work also in some other projects including teaching Blender in Universities , Blender Days in Brasil, Blender Pro (Brazilian Blender Conference), and so on :). Well, I am a reader of the anandtech.com website which is also a good place to see tests of servers, workstations, Graphic cards, SSD, processors, coolers, etc.
They started to use Blender as a benchmark tool a few months ago, and due to the transition to 2.5, they have stopped to use Blender until it’s final production launch. I’ve made a complaint about it, because Blender is one of the few software programs that is sufficiently stable and multi-threaded to test super servers, even in it’s Alpha Stage. (64 Threads, 64GB RAM in 64 Bits O.S.)
They asked support to test Blender in a Bulldozer Instanbull Opteron, offering an e-mail to contact Johan at anandtech.com, great person ! After 3 e-mails, I explained how to use Tiles and Threads in Blender 2.5 Alpha 2, and the test was made in a Dual Twelve core Opteron Server, with a ”Robot Scene”.
Click here for Anandtech test results [fixed link 03-30-10 1019]
link’s dead and I can’T find any traces on the homepage either.
“64 Threads, 64GB RAM in 64 Bits O.S.” I want one :-)
Hi Ivan, nice tests.
two things:
1) the link for Anandtech test seems offline. Is this the “blendernation” effect? Or the link needs fix?
2) I found the graphics confuse :) If the idea is to compare the machines and the operational systems I think they could have put more time on that (or that’s probably in the page I can’t open ;)
Once again interesting results. Is there any explanations why there are so many variations between the results in Linux and Windows?
http://anandtech.com/show/2978/amd-s-12-core-magny-cours-opteron-6174-vs-intel-s-6-core-xeon/7
Interesting bench but is there anywhere on the web we could have a graphic-cards blender benchmark, gaming and pro.
I have not seen any new blender graphic test since the BBB’s one, and with all the features involving graphic that Blender has, mainly polysculpt, it would be interesting to know what cards are the best to invest in nowadays in order to have the most comfortable performances while working in the view-ports.
If you have ever seen such a benchmark please let me know, thanks!
To me the surprising thing is how much speedup they get out of Linux. Can’t wait to have the same performance on the Windows side.
Luckily they did mention that in fact the software is still in Alpha hence the results should be takes with a grain of Salt.
Either way, good start.
Yay. linux beat windows.
Another proof of superiority of Linux ;)
Or at least the Linux version of Blender :)
Comparing the windows and linux versions is useless here as it’s not said which compilers were used and with which options. (or I’m blind)
Hi dfelinto, and everyone at B.N.
1) Anandtech website change it’s visual style and then some page reallocation ocurred during this transition. (3 days ago)
2) The graphics seems confuse at fisrt sight, sometimes we have to read full story articles to understand what’s being done. Course, I will continue supporting them to do better tests using Blender as Benchmark tool.
The explanation of the variations of the Tests:
For Windows:
Windows Vista/7 uses a new kernel that is more preemptively scheduled than the old multitask kernel used in Win2000/XP/2003. Then the tasks are evenly distributed over the cores ( changing threads over time from processor to processor) . What’s happened to Intel processors ?
The kernel changes threads over processors including HT ones that aren’t real ones. My (later) suggestion to Johan was to disable HT in order to use only Phisycal processors, or set affinity manually, to improve test reliability.
In Linux:
Linux is a multitask kernel that does not change threads from processor to processor over time, and then, when a non Physical processor is found, the thread is tied to this processor until the task is finnaly done, but in this case, performance problems in HT (non Physical cores) may happen.
The test made, didn’t take into account those differences, and it’s not compiled to take advantage of some instructions. I’ve already sent an e-mail to Johan explaining those HT problems, processor scheduling problems and some more hints. That’s the only way we can benefit from those tests before a decision to buy one or another Render Server. There is no Biased tests, the problem is that the setup needs to be carefully made to determine better results.
Ajusting Blender to fullfill a great number of processors is also another point, because some tiles are more demanding of processor power than others, and then, a Tile Preset for Threads seems impossible sometimes. AND, the eofw scene test is reaching its limit.
I’ts a good work to research more and more this Area. It’s helpfull to persons that need to accomplish great Jobs with few Machines, Render Farmers, and even Gamers/Overclockers.
Good Vibes !
Thanks for that explanation.
Another way is to either make more tiles. And increase threads beyond the limit of the CPU.
For example. In the old 2.49, on my quad core, I still select 8 threads. This delivers maximum performance. As threads are continuously occupied with more then a single tile. So if one finishes, there is no wait for the next.
On a 2x 12core MangyCore, giving it 64 threads might deliver maximum performance as no core will go idle when one of the tiles is finishes.
As for tiles, that indeed will vary.
Just my 2c
Thanks Ivan, that was super informative!
So Linux is better than Windows? That cannot be inferred from this benchmark. What Windows Version+Patchlevel was used? What Linux Kernel? What graphics drivers? What compilers are what version? Maybe it just means that the Linux admin is better than the Windows admin. Or the compiler used under Linux is better than the one used under Windows. Or that blender is better optimized to Linux. Or, or, or. Thats the problem with benchmarks, there are so many variables. To get a proper answer to what is better for rendering stuff (Linux or Windows) all these points have to be varied (compilers, renderers, OS versions and even admins).
What would be more interesting would be a comparison between Blender 2.49a and 2.50 Alpha 2. Both compiled by the same admin with the same compiler and run on the same system using the exact same .blend scene. The things can still be varied in order to see how much they influence blender.
Blender has always run faster in Linux than Windows. I’m not at all surprised.
@panzi, first off, the graphics driver makes no difference in renders. Second most of what you are talking about is covered in the anandtech article and by the poster of this article.
There is a big difference in the way that scheduling is done in Windows 7 and in the Linux kernel. The blender rendering engine seems to work better with the the scheduler in Linux than in Windows. It may be possible to override the scheduler in windows 7, but that would have to be done in code most likely. I’m sure that if you did the same render in Windows XP the results would be much closer because windows 7 uses a different scheduler.
The reality is though if you are looking to build a render farm you aren’t even considering using windows 7 unless you have waaaay to much money lying around.
@panzi
As I said either Linux is faster OR the Linux version of Blender is – that much can be inferred.
I’m no expert, but I don’t believe that all those minor variables would have contributed to such a BIG difference. Of course it may be that it’s just Windows 7, but then what’s the point of using an obsolete system (XP) just to run one application faster?
Anyway, hopefully we’ll get a chance to see the results when the benchmark is more optimized. Whatever the “reasons” for the difference, I just wanted to feel good about the fact that I’m using the faster option :D
All this bla,bla, translate into a pumping iron CPU computer to run blender 2.60- Right ?
(I just try to make constructive criticism, I hope that’s clear.)
@deathguppie Currently the graphics driver might make no difference, but I think in future more and more technologies like GLSL or OpenCL will be used. Then it very much makes a difference.
Oh and things like compilers and their versions might make a big difference if they support things like OpenMP (more or less transparent code parallelization).
I think i should consider buying a new rendermachine, as my current one takes over 17 minutes for this scene with the described settings :-P
Interesting, but they are obviously not using optimized builds on either platform (I get similar Linux times on a single i7).
My pc took 1 mins 3 sec to render this pic. Render was set as described in the benchmark.
OS : Windows 7 64 bit
Processor : intel i7 860
Ram : 2 GB DDR3 (bad)
Graphics Card: nVidia GT 9600 512MB (bad)
waiting for a 8GB single slot ram and for upcoming nvidia quadros and ati firepros :D.
Blender needs CUDA. Once this is implemented, we will see a big jump towards epic pictures.
Hay My Kubuntu 9.10 64 bit rocks!
43 seconds only. Far better than windows (1min 3 sec).
I think i shoould install official driver from nvidia for my Kubuntu to make it faster.
linux beats windows yahoooooooooooo!
I wonder if there is a bench mark for an i7 64bit processor
What?? Linux is faster than Windows?? NEVER!!
Did you really read the results??
The results are not sorted by CPU but by render-time!!
I use a 2CPU XEON-System (2.5GHz) with in summary 8-cores.
My result is 36.26 seconds!!
No reason as Blender-user to buy this expensive CPUs!!
Correction: I use an up-to-date SVN!
But nevertheless …
Update: my Cinebench 11.5 result is 6!!!