Render Engine Comparison [Methods and Technologies]


(zeealpal) #1

Here is a scene I have prepared to allow the comparison of how different render engines handle different light situations. This will allow you to see the limitations of the methods each engine uses, as well as its strengths. This does NOT compare how they handle materials, or which materials they have, or what is better nodes etc… please leave that out of this.

It has a small light source, reflective caustics off a mirror, refractive caustics through glass, and the dreaded caustics viewed in a mirror as well.

SLG3 Using BiDirectional Path Tracing + Vertex Connection and Merging with max 12 bounces. 3 Minutes/162 passes Intel Core i5 750 Quad Core 3.3 Ghz.


Please render at 960 * 540 resolution, and please keep the render times around 3 minutes (longer if your machine is significantly older/less powerful)!!. Quote your hardware as well as the engine you used, as well as any settings, like my example above.

The scene has materials with names showing what that material should be. Please try balance the lighting to the image I posted as reference.
.Blend here BlankComparisonScene.blend (99.8 KB)
Here is the floor texture. http://cgtextures.com/texview.php?id=33550
Here is the HDRi. http://www.openfootage.net/?p=355

The engines I hope people will test include (but not limited to) Cycles, LuxRender, Yafaray, SmallLuxGPU, Mitsuba, V-Ray, Octane any any others you wish.


(mib2berlin) #2

Ha, you change the blend. :slight_smile:
Octane 1.0 RC3
Pathtracing with Monte-Carlo-Integration
Maxdepth 8
Caustic Blur 0.01 (Clamp?)
600 S/Pixel
3 Minutes
GTX 550Ti


I think it is a good Result for a brut(al)e Pathtracer on a low end GFX card.

Cheers, mib.


(zeealpal) #3

That’s quite an impressive render, I do like the way the mirror turned out so well :slight_smile: quite a good result


(J_the_Ninja) #4

Big Lux:

Defaults (bidir/MLT, 49 s/px)


SPPM (57 passes):


2.8Ghz Xeon W3530


(bashi) #5

trying it with mitsuba. :wink: photon mapper fails (Irradiance Chache don’t work somehow?), Only Primary Sample Space MLT gives usable result so far:

Integrator: Primary Sample Space MLT
Direct Samples: 48
Samples: 512
Pathdepth: -1
Russian Roulette: 12
Luminance Samples: 40000
Biderectional: Yes
Two-Stage MLT: No

i7 2.6 (macbook), 8GB Ram, osx 10.8
rendertime: 7min 33sec


(mib2berlin) #6

Please render at 960 * 540 resolution, and please keep the render times around 3 minutes (longer if your machine is significantly older/less powerful)!!.

Hi bashi, it is hard to compare render engines when you render double as long and with half size.
Please add also your system specs.
As far I can see it is very clear in foreground but not in mirror, hm.
I never test Mitsuba because can´t built it on my system but follow the development still. :slight_smile:

@J_the_Ninja, do you have a very glossy gound? It is also very bright.

Thanks, very interessting.
Mib.


(Repgahroll) #7

Caustics aren’t the strongest point of V-ray, but I attached a V-ray version for comparison, 2’10" on a mobile Core i5. Windows 64-bit. Light cache depth: 100. Could’ve increased the caustics samples though.
I have a friend working for an animation studio and I’m gonna ask him to render in Arnold if possible.
And when I say V-ray kicks serious arses you don’t believe me.



(bashi) #8

@mib2berlin: you’re right. i was a bit in a hurry…

here Mitsuba Photon Mapper for now:

Blender own osx build r52438
Renderer: Mitsuba 0.4.2
Integrator: Photon Mapper
Direct Samples 2, Glossy Sampl. 2, Max Path depth 12, Russian Roulette 5, Global Photons 250000, Caustic Photons 400000, Volume Photons 0, Irradiance Cache: Final Gather Res 14, Overture, Quality 0.8, Irradiance gradients, Neigbour Clamp, Quality Adj. 0.5, Sampler Low discrepancy, Pixel Samples 30
(Those are the settings that are changed)
Render time 3min 2sec (on i7 2.6ghz, macbook pro 6.2, 8GB ram, osx 10.8)


Note: The Floor Material is Diffuse only. haven’t found a good way yet to mix glossy to it, party cause mitsuba doesn’t know glossy, but instead Smooth Plastic or Metals… material setup for mitsuba not really most intuitive, have to make some more test to figure that out :wink: Mirror is Smooth Conductor (Metal) - Silver
I had to replace the lightmesh with pointlight… mesh emiiter gave fireflies only, almost.

there are still some issues, like behind the mirror in the mirror the bright spot.


(mib2berlin) #9

@Repgahroll, looks really good, in case of your system and time seams the best.
@bashi, cool, nearly noise free but caustics looks better with your first render.
Has Mitsuba not also VCM like SLG3, as far as I know the SLG3 VCM comes partly from Mitsuba.

First try with Cycles, it seams it is not possible with Cycles to get it because of Pathtraycing only atm.
Does not matter point light or mesh emitter.
Full GI
1500 Samples
3.27 Minutes
GPU
Specs in my signature.


Next try is Indigo. :slight_smile:

Cheers, mib.
EDIT: Cant get anything useful out of Indigo, my fault for sure.


(Dade) #10

Nope, SLG3 VCM comes from SmallVCM: http://www.smallvcm.com/


(Tame) #11

Here are my results with Lux :slight_smile:

Luxrender isn’t really suited for fast caustics rendering such as these… I tried to tweak the settings so that the caustics would show in the mirrors in 3 mins…


Just for comparison, if you let it render longer it will eventually start pick them up, this is 15min render.


Then SLG3

Old Path GPU, nice 10150k samples/s but it doesn’t look that good



(Tame) #12

Path CPU, I think it looks better than the GPU


Bidir CPU, might look the best if you let it render longer


Bidir VM CPU, the fastest render method to get clean image quickly



(Repgahroll) #13

Okay, here is another V-ray version, a little bit noisy and blurry, I don’t have experience with caustics in V-ray, for product visualization I’d use an unbiased renderer. Anyway, 1’70" on the same mobile i5 processor. It’s using full IM+LC GI although in this scene it’s hardly needed, if I disable them V-ray will put the competition to shame, but it wouldn’t be fair.



The SLG3 versions are better imho, the problem is that it’s fast only with very simple scenes, I’ve tried rendering the classroom scene and even after some hours (rendering on mode “6”), the scene was still very noisy.


(m9105826) #14

If no one gets to it before me, I will run this through Arnold if people would like. Be aware that Arnold doesn’t currently handle any kind of caustics though, so it would be a bit of a waste on a scene like this.


(mib2berlin) #15

@Tame, Luxrender is slow but looks good.
@Repgahroll, your first render looks much better, its like postpro glow/blur effect in the second.
Nobody has to shame or be proud for an engine other people develop, get the best out of the engine in 3 Minutes. :slight_smile:
@m9105826, it would be very interesting to see a render from Arnold.

Cheers, mib.


(zeealpal) #16

I’d hardly call it.unfair, of the result is comparable to.the other engines without full gi, then that is great for vray.


(m4x3d90) #17

Hello everyone!

I’m looking forward to challenges like this. They are good opportunity to learn and tweak stuff.

Here is my frame rendered in v(-)ray. As seen no background is added due to mine small knowledge about this app. So I would be pleased to get new info.

I think it’s hard to compare that render to others, because that one makes every stage on separate pass. So Actual time, spent for rendering follows.


RenderTime: 1m 44.2s
Windows8 64bit
CPU: Intel i5-2430m 2.79GHz
4GB RAM


Building caustics photon map…: done [ 0h 1m 50.0s]
Building light cache…: done [ 0h 4m 54.7s]
Prepass 1 of 1…: done [ 0h 1m 38.3s]
Rendering image…: done [ 0h 1m 37.8s]


Number of raycasts: 87652309
Camera rays: 5151961
Shadow rays: 7445614
GI rays: 5898810
Reflection rays: 40976442
Refraction rays: 4709981
Unshaded rays: 1868985