How does your system compare to other systems when it comes to rendering your scenes in Blender? What's the best system configuration to optimize your rendering times? These questions, and more, have all crossed our minds at least once. Well, here's a way to begin to get some real-world information towards answering your questions.
Jarod has recently updated his render benchmark webpage. He writes:
Now it's easy to submit render values. There's improved clearness in the results including ascending and descending sorting.
He's made it quite easy! Simply
- Download the .blend file
- Render the scene
- Return to the webpage and register your system details and render specifics using a nicely formatted series of forms.
There's lots of details to study so you can get your own system optimized or, it can help you to make a more informed decision concerning system upgrades. Thanks Jarod!
Check it out, here.
42 Comments
Yeah! It's a good question.
Thank for the info :-)
Testing... Testing... Testing... Testing... Testing... Testing... Testing... Testing... Testing...
Cool.
I'd also like to see something like this with gfx cards problems and solutions :)
Why aren't I surprised to see unoptimized linux boxes crawling behind windows machines...
But to be fair, someone should do the same benchmark with a similar machine running Linux. If one can set up something like that that is... installing video card drivers, downloading missing libs etc... aaaah the fun of Linux :D
No Mac OS X results right now?
Nice how one can click on the tabs to sort that colom.
Seems to me the windows in current list is faster because that cpu is faster, so yes if you would want to know which o.s. is faster you need to install all operating systems on the same machine,... but I didn't understood this to be an o.s. comparison.
Well OS architecture can modify speed. Although it's never something dramatic. Just shear curiosity :D
A linux renderfarm would pulverize the score no doubt :D
nice idea and work so far, thanks for sharing! It took me a while to realize the form wasn't working because I had referrers off.
(perhaps: Info collection and website submission could be built in the .blend
http://docs.python.org/lib/module-sys.html sys.platform
it could generate a url, then the site would ask info the .blend couldn't collect, verify info and captcha on the site before final submission.)
Excellent, I was thinking in a little upgrade to my computer, and this could be very, helpful for me
thanks, Jarold!
aws357 :
windows xp pro sp2: default settings time: 1m 54.13 sec
Linux FC6 kernel 2.6.19 : default settings : 1m 18.74 sec
linux if faster by about 31% quite a big difference.
Vanilla blender 2.43.
Hello
Thanks a lot for submitting your values!
How you could use the data from Blender Render Benchmark?
First there is no warranty, if the values are really true!
But there is a simple way to check trustability.
Sort by i.e. CPU, if the same CPU listed several times and the render times are nearly the same, so it's seems seriously.
You can also see how the CPU performs with other Operating systems, if different OS are listed.
So the valuable information isn't the fastest CPU in list.
When you look at the list, it's easy to find out...
...that Linux is mostly a bit faster than Win.(at the same CPU)
...that Win Vista doesn't kill this much performance compared to WinXP.
...the more Cores/CPU's the better.
...that optimized builds (i.e. SSE2) could increase render speed definitively.
...that bigger CPU cache memory isn't an amazing speed up.
...that MHz isn't MHz.
...if Moore's Law is right? in 1996 Pentium 200Mhz 1:23:03.46 vs. 2007 Intel Core2Quad ~10sec?
in my opinion a QuadCore CPU running under Linux with optimized Blender build and render via command line (no xserver) should give the fastest results.
thanks over again!
Jarod
@aws357:
Here are some results from an Apple iMAc 24 2.16 Ghz 2 Gb ram with Blender 2.43
x:4 y:4 threads 1 02.24.16
x:4 y:4 threads 4 01:14.60
x:4 y:4 threads 8 01:13.20
OOOOOH! YEAAAH! BABY...
80 CORES!
http://news.com.com/Intel+shows+off+80-core+processor/2100-1006_3-6158181.html?tag=newsmap
Calcule now.
XD
5:40:08
that's no good. And always that reminder that my CPU gets 70 degrees celsius.
Maybe more RAM and beter cooling...........
Using Blender 2.43, WinXP on a P4 2800 MHz w/ hyperthreading, 1 thread is 4:26.14 and 2 threads is 3:47.61. I thought it would be a lot closer to a factor of 2 improvement, because the cpu usage meter went from 50% to 100%. Is it memory bandwidth limited (I have 2 GB ram)? But even so, it seems to me the Windows CPU meter must be inaccurate, then.
I tried the tool this morning in the office and noticed that at times it just skips pages. It also happens that it completes a form with genuine data although I didn't enter it, nor is it applicable to my machine ??
I tried my Intel laptop and both memory and cpu got 'changed' a few times ... it also tends to skip from page 2 to 7. Clicking back makes the 'fake entries' visible.
Just checked with my home pc and I have the same result. Mind you, I'm running Firefox, didn't try IE.
can concurrent entries to the webform get mixed somehow, leaving me with the entries of another blenderhead?
yep same problem as monki
this page works horribly with firefox
a few years ago it was understandable
spent 20 mins just trying to submit the damn data
and i refuse to use internet exploder, i prefer to browse the internet virus free :P
i give this site a 10 for idea, and a -10 for execution
cheers
@monki and wasamonkey (those two names work pretty well together LOL)
Try installing the Firefox IE Tab extension. It allows, via a simple statusbar button, any tab to be viewed using the Firefox or Internet Explorer browser engine. This extension really does help with a lot of websites.
@ JOhn B, remember that HT isn't the same as dualcore, only dualcore will give near-2x performance boost.
The CAPTCHA subjugates me completely! Apparently I am a computer.
Debian Sarge, 2.6.16.41
Intel Pentium 4 2.8GHz, 512K L2, HT
Intel 865PE chipset, 1.0GB RAM
Blender 2.43
1 thread: 4:15.34
2 threads: 3:35.36
@monki and wasamonkey: sorry for the troubles, but I've tested the form on 5 different PC all under firefox and works fine.
I' also use just firefox for web browsing.
Hmmm, must be a problem whit php sessions, anybody out there who has the same problem?
please contact me by email (address found in blend file or at website)
Thx Jarod
Most people seem to stop at 2 threads. Crank it up to 8 threads to see what you get. I know that the threads setting is "supposed" to align with the number of CPUs or cores in a computer but, I have a dualcore CPU and I have done many, many, many render speed tests and my fastest render times have ALWAYS come from using:
8 threads
20x20 parts
I don't care about what's "supposed" to be best, I just know that I ALWAYS get my fastest render times using those settings. Your computer might work best at different settings but, my point is to suggest that people experiment a little bit more.
Also, looking at the data grid, I see that most of the tests that used 8 threads reside among the fastest render times documented...hmm.
I had the same problem, it was skipping pages and when it finally got to the summary page it had all the wrong choices listed.
I use Opera.
I'm not sure if it's that great that the "here" link on this page has a php session ID in it?
@alexander
It's a link that I just copy/pasted. It seems to go to the same page regardless so, I removed it.
Thanks, I guess. :) I have no idea what the problem is with a php session ID but, it's gone. Maybe there's a security issue?
Well finaly got around to test the test file :)
I tried various configurations including 8thread render... and intresting results.
X2 3800+ @2.6ghz
4gb ddr 400
DFI Lanparty SLI-D
Blender 2.43 Vanila
Threads Tiles Time
2 4x4 1:36.86
8 4x4 1:32.74
8 20x20 1:30.69
Blender 2.43 sse2 (primary version i use)
Threads Tiles Time
1 4x4 2:20.71
2 4x4
Well finaly got around to test the test file :)
I tried various configurations including 8thread render... and intresting results.
X2 3800+ @2.6ghz
4gb ddr 400
DFI Lanparty SLI-D
Blender 2.43 Vanila
Threads Tiles Time
2 4x4 1:36.86
8 4x4 1:32.74
8 20x20 1:30.69
Blender 2.43 sse2 (primary version i use)
Threads Tiles Time
1 4x4 2:20.71
2 4x4 1:16.01
2 20x20 1:12.50
4 4x4 1:15.56
8 4x4 1:13:12
8 2x4 1:14:45
8 20x20 1:10:84
Blender 2.42a 64bit
Threads Tiles Time
1 4x4 2:18.77
2 4x4 1:13.66
2 13x10 1:11.17
Blender 2.43 sse2 CPU at 2.7Ghz (.1ghz faster then above)
Threads Tiles Time
8 13x10 1:08.33
During the experiments i noticed one thing. The 20x20 tiles don't result in 400 tiles, but in 130... 130 seems to be Blenders maximum, at lest in the three versions tested above. no matter what i set the x and y tiles to (above 13x10) result is always 130 tiles... :(
The 64bit version only has option for threaded or non threaded. so testing 8x threads was not possible.
So in the end the "8" threads and 13x10 tiles actual help :)
I agree
Can anyone explain this to me?
On my friends MacBook, we performed the test with the native OSX version and a WinXP running on Parallels Desktop emulation:
* Rendertime on the Mac native version: approx. 2:30 min
* Rendertime on the Win version: approx. 0:45 min
I know that the hardware specs are much better than a 'normal' Wintel duo core laptop but still ... I expected emulation to have a little drawback on performance. We actually only allowed Parallels to access only 256 Mb RAM. Rendertimes dropped to 1:20 when we added more memory and THAT is weird IMO.
Maybe the OSX version of Blender needs better recompiling ...
Any other thoughts ... and yes, we tried it several times and my glasses are up to spec :-)
To be honest, the page should be structured in a searchable database, to be really useful.
Having a big "monolithic" page will be very inappropriate when the number of entries will grow considerably.
Good initiative.
What's up with the high scores from the MacBooks with the Intel GMA 950 leech graphics? I was holding off to buy a current Mac because of the "crappy" GPUs. But the Core 2 Duo MacBooks have beaten many machines with comparable processors and dedicated GPUs. They've even beaten MacBook Pros with dedicated ATI Mobility Radeon X1600s and the same OS versions and amounts of RAM (and even *more* VRAM). I don't get it. Is it because the GMA 950's drivers are open source and Blender is somehow optimized for them?
A Mac Mini Core Duo with a GMA 950 did pretty well, also. I was thinking of getting an older Mini G4 with a dedicated Radeon 9200 because I thought it might do better with Blender, but that one is way down on the list while the GMA 950 one is towards the top. Hmm.
@Ivor
I haven't looked at those results but, you might want to ensure it's a fair comparison by checking the settings they used as well (if you haven't already done that).
Yes, I'd say it's definitely a fair comparison. I checked the specs on the chart (what few there are) and looked up the specs of the MacBooks (on lowendmac.com). The plain MacBooks are doing incredibly well with this render. Perhaps they are doing very well at this task, but wouldn't on others. I wonder how thorough this render is at testing the various processes of a GPU. I can't do hands-on to find out myself because my computer is currently sick with capacitor plague.
The first MacBook (not Pro) comes in at number 98 at 00:01:26.48. It has a Core 2 Duo running at 2000MHz. It's doing 2 threads. It has 1024MB PC2-5300 DDR2 RAM. It runs Mac OSX 10.4.8 and Blender 2.43. It's coming in at number 98 out of 846. A lot of these other systems are real BFGs, too, and the MacBook is holding its own among them. The second MacBook is a similarly configured Core Duo and comes in at number 110. Their GPUs are the supposedly pathetic Intel GMA 950 leeches.
I'm just wondering how this can be. So far it's changed my mind about wanting a current MacBook. My main reason for my next computer purchase is Blender. Did Apple work with Intel to super fine tune the GMA 950 so that it's actually decent? It seems to be a real anomaly. Go ahead, check it out for yourself.
Still hope I'm not going to kick myself in the ass for buying a MacBook as a Blender box. Maybe I should wait to see if Jobs will cough up some decent new systems in June.
Well, I'm a PC user so I really know nothing about Macs. Another thing that you might want to consider is that most utilities, patches, etc. that are made available seem to be compatible in the following order of priority:
PC/Linux
Mac
That's not always true but, it's the pattern that seems to predominate. IOW, I don't think I've ever seen something developed with Mac compatibility as the first priority. Eventually, but not first.
I'm just a noob and I've been trying very hard to figure out which computer to buy to do Blender. What I've found out is that there's no way I'll get a MacBook in their current state. Various benchmark tests at barefeats.com have convinced me of that. I didn't realize that rendering depended more upon the processor than the graphics card. The MacBook's Core 2 Duo is excellent, but the Intel GMA 950 GPU really bites. Sure a MacBook can render, but for other graphics intensive uses, forget it.
I think the best option for someone like me who is on a super-tight budget is to build your own PC out of parts from newegg.com and such places. You can build a great Core 2 Duo computer with *two* nice graphics cards running together in SLI mode for a less than a MacBook. And you can get a free *nix OS like Ubuntu. Seems like Blender would be much happier in that environment.
Somewhat unfortunately for me I have Final Cut Pro so I have to have a Mac. Macs are great, but Apple isn't currenly offering the "prosumer" anything decent in the way of GPUs. I just bought an older 1.33GHz dual G4 an hour ago. No screaming SLI for me, but at least I'll have a Radeon 9800 Pro (hope Blender and Ubuntu like it).
This benchmark test doesn't use the GPU of you graphics card at all (you would need something like Gelato for that). So only 100% CPU and memory test, that is if you have enough ram and don't have to use the harddisk. I'm using a HP xw4400 workstation (E6600 cpu and 2 Gig ram) and have a time of 1 min with 2.43 SSE2. Have done the same test with a Mac Pro (2 dual core Xeon 2.66 GHz and 5 G ram) and found 30 sec needed for the bench. Was thinking a upgrading to a Quad core QX6700 cpu on my HP but now I have heard Mac is comming with a 2 x quad core 3GHz Xeon ! Don't think this can be beaten with any single cpu machine ?
Where are the Pentium M's in the list???? What about all the centrino Notebooks of the last 5 years??!?!
Does the graphic card is not important in the benchmark?????!!!
The benchmark is about rendertime, not realtime onscreen drawing. So no, the graphic card is not important for this benchmark.
So, correct me if I am wrong: it is not important what graphisc card you have when it comes to make animations?
Who says this benchmark is about making animations?
It's about rendertimes and for rendertimes the graphics card is un-important.
If you want a benchmark on graphics cards you are free to start your own website with your own specs of importance.
But to answer your question; no the graphics card, or any hardware for that matter, are the most important when it comes to making animations: It's creativity and that does not come out of a box.
Well, as of today (June 16, 2008) my Xserve G4 Dual 1.33 GHz is average in the benchmark. It is number 1027 of 2055. My 5 year, 4 month old computer is still holding its own at the midway mark. Not bad! Not bad at all! :)
I'm very grateful to the Blender developers who are continuing to support the PowerPC platform. There are millions of cheap old Macs out there, and folks who can't afford new hardware are still able to educate themselves thanks to the continued support.
Just did a test on my Precision 1500
Intel® Core™i7 2930MHz 8192MB memory
Win Vista 64Bit 2.49 64bit
Result 00:00:23.15
Linux Fedora 12 64 bit
Result 00:00:21:79 (Not published)
The linux system was run out of the box, but with nvidia driver. Exactly the same hardware.
Have not tested on Windows 7 yet, but will try that as well.
Regards Paul.