BlenderMark. A very simple benchmark for everyone =)

Hi Blenderpeople!!

I wanted to create a quick and simple benchmark for blender that everybody could compare results of rendertime and system secs.

Im not a blender pro, but i thought about what the bench should do:

  • Quick and easy to build
  • Massive CPU loadmuch reflections and shadow calculation, some procedural textures,no image maps.
  • Dont use too much ramit should not blow up the system and use GB’s of ram it should stay below 64 or 128mb cause the amount is only important for complex scenes not for plain render time.The access speed and transferrate of the ram is benched anyway if not everything fits in the processors cache.
  • Simple geometryThe .blend file should be somewhere around 1mb to allow quick and easy sharing and whatever :slight_smile:
  • Rendertime should not be to lowif it renders in one or two minutes the next generation of 4x4 CPUs would do it in some seconds and the gaps between the different systems is to low for a good compare. And everybody likes to see his CPUs work a bit :wink:

Okay with these things in mind ive done a first approach to discuss here and to improve the whole thing. Ive put the benchmark on my small site.

http://www.pixeltrap.de/downloads.html

Everybody is invited to download it and if he likes to improve it, modify it or whatever u can post improvements here or send me an email. The adress is found on my site. To render, just download the file and press F12 in blender. U can post ur system specs and rendertimes here as well =)

My Specs:
Rendertime: 19:06.96
Blender version: linux2 dynamic build 2.41
threads enabled: yes
CPU: AMD Athlon™ 64 X2 Dual Core Processor 3800+
MEM: 2048 DDR3200
OS: Linux version 2.6.16.20 on Ubuntu dapper drake 6.06
Board: Asus A8N5X nforce4

maybe i will create a small site where the fastest systems with some more
detailed specs are shown so somebody can look what cpu is a good deal
and what OS and so on…

sorry for any bad spelling… cause english is not my
native language, hope u dont mind :slight_smile:

Thanks to everybody for reading and for the interest!!!

http://www.pixeltrap.de/dnl/Blendermark01_19_06_96.jpg

20 minutes running time for a benchmark (running on a fast processor) is waaay too much.

okay…
but why?
I agree that if you would run it on a very slow CPU it would take hours.
and on some faster CPUs maybe it would finish in about 10 minutes.

I could reduce the ray shadow complexity, that it would run in about 4-5 minutes
but i think that faster systems (maybe in the future) are too fast to
compare accurately with this reduction

what could be the best compromise?
what do others think?

thanks,

There are two reasons one runs a benchmark:

  1. to see how fast his system is compared to the others. Let’s say one
    just found this benchmark (me in this case) and wants to try it. But when
    the benchmarks will run for 30 minutes (that’s wasted time, because you cannot do anything else in this time, that’s the way the benchmarks work) the use will think twice - can he afford to loose 30 minutes ?
  2. to optimize blender speed - make some modifications, run the bencmark and see if it works. Waiting 30 minutes to validate your one line of code modification is too much.

I would say it is acceptable for a benchmark to run for 3 to 5 minutes on a slow system.

You can find a Blender benchmark at http://www.eofw.org/eofw/. Might be worth checking out.

Good luck on your effort though.

hello BeBraw!
That was exactly what i had in my mind…
Maybe i make a another approach like this,
but if somebody has done it already… why do it?!
ill think about it…

@cipix i give u right 20 minutes are too much
(not too much, but too much for people with slower ones)
Ill reduce the render time without losing quality :slight_smile:
5-6 shadow samples are enough anyway :slight_smile:

thanks for all comments so far