• Monday , 1 June 2020

Research ARNOLD vs. CYCLES

Code Canyon


Maya 2018.2 with Arnold 5
Blender 2.79 Daily Build

DELL XPS 15 9560
i7-7700HQ @ 2.80 GHz (4 cores / 8 proc)
GTX 1050 (4 GB)

Latest Win 10 / latest Nvidia driver

The still life scene was originally set up in Blender 2.79 with photogrammetry models by Oliver Harries:

Special thanks to Oliver Harries for these fantastic free and very detailed models.

All models will use a color, roughness and normal map. All maps were scaled down from 8K to 2K for this test.

Lighting with one High Dynamic Range image.

The Kiwi’s fur has been converted into meshes for this test and used for both scenes.

Both scenes will use Physically Based Rendering by aiStandard (Arnold) and Principled (Cycles) and 64 samples for Camera (AA) and all other passes (Branched Path Tracing in Blender).

The video shows the different render times in real time.

Top left: ARNOLD CPU
Top right: Cycles CPU
Bottom left: Cycles GPU
Bottom right: Cycles Hybrid (CPU + GPU)

Render time might be affected by recoding with OBS Studio.

In this example ARNOLD and CYCLES are very equal in CPU render time and quality. CYCLES is up to 20% faster. ARNOLD has less noise especially in the darker parts. The Principled shader in CYCLES is using Multiscatter GGX what will result some areas with brighter highlights. ARNOLD just supports GGX. Some parts like the wooden bowl will be lighter. Beside those differences the quality is very similar.


3d Ocean

Related Posts


  1. I Commando
    December 26, 2018 at 00:36

    I think that people are missing the point and are just looking at the fastest.
    By a large margin did Arnold handle the shading in a much more close to real life way.
    Look at the kiwi. It looks so much more realistic and that is not something easy fixable to do in Cycles.

    Both have their strengths and weaknesses. But for commercial work where color and realism are key I would pick Arnold.

  2. JC
    December 26, 2018 at 00:36

    What Arnold settings are you using? Samples, Ray depth, etc.

  3. v n
    December 26, 2018 at 00:36

    VERY IMPORTANT for CYCLES GPU: Set cluster size to 256*256 pixels, it is much much faster than default !

  4. Moon Base
    December 26, 2018 at 00:36

    Is it just me or does arnold change the colors of things?

  5. MonsieurCoinCoin
    December 26, 2018 at 00:36

    Sadly, hybrid isn't in the latest stable blender release :<
    I tested it in it's firsts daily builds tho, it crashed a few time back then but it was faster ♥

  6. Juan Carlos Gutiérrez
    December 26, 2018 at 00:36

    Hello there. Do you mind sharing the already setup scene (not models) in order to make a comparisson in appleseed 1.9 render? Or if you can do it, it'd be great!

  7. Michal Rechtorik
    December 26, 2018 at 00:36

    GPU could be much faster if you used bigger tiles.

  8. Nizar Amous
    December 26, 2018 at 00:36

    same scene in redshift please ? 😀

  9. Invertex
    December 26, 2018 at 00:36

    Now do a test with Redshift and watch as it blows all these away 🙂

  10. Tanmay Khelkar
    December 26, 2018 at 00:36

    Fuck, this is disturbing. I was planning on purchasing Arnold (for Maya). You saved me, man.

  11. John Draisey
    December 26, 2018 at 00:36

    Nice comparison. For the next test, you should try a few dozen light sources and really high ray depth and glass materials.

  12. Julian Morrisco
    December 26, 2018 at 00:36

    I'm sure you have plenty to do, but I've heard Arnold is at it's best (compared to other renderers) with multiple light sources, in fact a lot of light sources (I've heard comments about hundreds or thousands in some interview with someone working on a feature film). Food for thought, hopefully. Naturally, I'm on team Cycles, given that I use Blender as well as Cycles for C4D.

  13. Stuntkoala
    December 26, 2018 at 00:36

    Nice comparison. Did you use the same ammount of GI bounces? Also you wrote Arnold had less noise, so wouldn't it be more accurate to compare the Rendertimes not after a fixed number of
    samples, but after reaching a fixed ammount of noise? Maybe the high
    frequencies in the image could be measured to check this (or maybe you can't and have to guess it – I don't know any tools to do this).

  14. matt xypnise
    December 26, 2018 at 00:36

    Hi, thanks for the test. Can you tell me how is your laptop acting in a long high GPU load? Have you experienced throttling?

    Thanks in advance

Leave A Comment

You must be logged in to post a comment.