Unbiased render engines speed test: Maxwell vs. Indigo vs. Kerkythea

[kwistenbiebel]

Hi all,
Since there are now three players in the 'unbiased' render market, I thought it was time to do a little comparative speed test.(Maxwell vs. Indigo MLT vs. Kerkythea MLT)

objective:
Comparison of output generated by three different 'unbiased' renderengines, all with equal render time (1 hour 40 minutes) and comparable setup.

Scene statistics:
- render time: 1 hour 40 minutes.
- image size: 1230 X 665 (downsized to be able to upload on PPB)
- Physical sky + sun. turbidity: 2
- camera: F:10 , shuttertime 1/125, filmspeed 800 iso
- Location Antwerp, 18 may 2007, 12.00 h.
- Materials: all materials are pure 'diffuse' . (= no reflection, bump, ...)
[NOTE]: the reflectance on the cello is fake (= part of the diffuse texture,
baked in)


Notes::
- Apparently the location and sun settings in sketchup are interpreted differently by each engine. The only engine that renders exactly the same sun position as on the SU screen is maxwell. I had to change time and date in Kerkythea (set 3 hours later, one month later) to obtain comparable sun direction. Indigo was slightly different but not more than an hour of.
- render engine settings:
Kerkythea: MLT (on pathtracing)
Indigo: MLT (on pathtracing). Bipt was turned off. Consec.Reys= 50
Maxwell: MLT based.
- Sky in Kerkythea: sun multiplier was set to 6
sky intensity was set to 3
- No post editing done on any picture.
- The perspective of the Maxwell image is slightly different than the others even though it was exported from the same 'page' in SU
- there is a surface missing on the left side causing light leak.


Here are the resulting pictures:

[kwistenbiebel]

1. Kerkythea:

[kwistenbiebel]

2. INDIGO

[note]: Indigo is the only engine in the test that generates 'Depth of field' automatically with these camera settings

[kwistenbiebel]

3. Maxwell

[kwistenbiebel]

I am curious about your reactions.

Regards,
Kwistenbiebel

[A_Minima]

I can't have any doubt about great rendering capabilities of Kwisten : and we meet us very often throught other forums. So, just about KT : it's not a real unbiased render engine. The MLT is faked and you can refer to this post in the indigo forum :
http://www2.indigorenderer.com/jooml...pic.php?t=1347
Kosmokrator : "...kt uses mlt system but its fake....i have a talk with kt writer and he told me about their plans and how work kt at the moment...the will try to make some real unbiased ver but in the middle of 2007 will be ready..."

The rendering time of KT is very interesting (regarding to the result, it's very very interesting) but the developement of the soft may increase the rendering time... even if MLT is fake : to fake is often a good method to simulate reality. The specular on the cello surface and other grain addition to the Kwisten's renders are here to prove it. So, biased or not, the render engine can't do everything for you

[kleinjakob]

Hey kwistenbiebel I think this is a great idea to compare them. My only concern is about the cam setup, since it seems they are really different. Thanks for the work!

Kerythea:

What I like most is the good quality of the shadows and the light/dark contrast, but on the other hand it shows the most noise.

Indigo:

As you wrote, the DOF is a great plus in this render and you can see the structure of the fibre on the bed best. It looks a bit washed out/contrast less. I think the color tone generated by the sun is great.

Maxwell:

Not to much noise, good diffuse shadows, but the cam differs the most from the other setups...

Overall: I like the Maxwell most, but I have to acknowledge that the chello looks best in Kerythea, since it is most plastic.

Well, just some things that come to my mind when I take a short glimpse at your renders.

Thanks Jakob

[jenujacob]

Originally Posted by A_Minima

I can't have any doubt about great rendering capabilities of Kwisten : and we meet us very often throught other forums. So, just about KT : it's not a real unbiased render engine. The MLT is faked and you can refer to this post in the indigo forum :
http://www2.indigorenderer.com/jooml...pic.php?t=1347
Kosmokrator : "...kt uses mlt system but its fake....i have a talk with kt writer and he told me about their plans and how work kt at the moment...the will try to make some real unbiased ver but in the middle of 2007 will be ready..."

The rendering time of KT is very interesting (regarding to the result, it's very very interesting) but the developement of the soft may increase the rendering time... even if MLT is fake : to fake is often a good method to simulate reality. The specular on the cello surface and other grain addition to the Kwisten's renders are here to prove it. So, biased or not, the render engine can't do everything for you

if u go deeper into the thread u will find that KT is indeed an unbiased render with actual MLT.. that discussion created quite a stir there!! it was all a misunderstanding on kosmokrators part..

-jj-

[A_Minima]

Oups !
You're right. The reason is I've read this post yesterday and when I saw the Kwisten thread, I remembered it and "cut-paste" this post here... So, Kwisten, I beg your pardon

[kwistenbiebel]

I believe all these engines are unbiased, but still have different looking output.
The difference probably is caused by the different tonemapping that is used.
Users of V-ray will understand that tonemapping can really make a difference in photorealism.

It would be too easy to say that a render engine that uses an unbiased calculation would be automatically physical correct.
There are more parameters to consider besides the mathematical algorithm used for computing light distribution.


Just as a reference:
Here is a rendering of the same scene but with a 'biased' render engine :
Podium 1.4.1 , using default sketchup sky, original image size 800X600.
render time: 29 minutes