Luxcore vs rest of the world

General project and community related discussions and offtopic threads.
daros
Posts: 223
Joined: Thu Dec 12, 2019 3:25 pm
Location: inside human language
Contact:

Luxcore vs rest of the world

Post by daros » Fri Sep 10, 2021 12:39 pm

Hi, this post is intended to do some justice to Luxcore's bidirectional rendering engine.

To that end, I've prepared a scene that is not the typical HDRI + toy car. It's not a very pretty scene, but it reveals a lot of interesting information.
It is a very easy scene to set up, without textures, (textures don't reveal much about the speed of a rendering engine), so that everyone can repeat the test effortlessly on other rendering engines.
However it is a scene that easily represents the complexity of a modern store or bar or lounge, with shiny materials, lots of glass, reflections, and so on.
I asked the people who did the testing for two results:
First result with a very fast basic set-up, without tweaking the rendering engine
The second result with an advanced set-up, tweaking the rendering engine to get the best. No baking allowed. If there is any precalculation this should be added to the rendering time.

But since the following results were not all made by me I am not sure of their correctness. I'm only sure of the Luxcore and Maxwell Render results that I did personally. So I really appreciate any corrections from you guys. On rendering engines like Vray and Maxwell I decided to publish the results only on GPU because these companies claim the unconditional speed increase on GPU compared to CPU.
As hardware I put the best hardware available today, which means the nvidia 3090 and the Ryzen 3990, neither of which I own nor the people who did the testing. I simply scaled the result based on the performance gap between the hardware used and the results of official benchmarks on similar rendering engines. This is one of the benchamark pages I used: https://www.cgdirector.com/blender-benc ... ed-scores/
For example if the test in Vray was done on a 2060, to get the approximate render time on the 3090, I divided it by 3.6, which is the speed multiplier I re-encountered.
I know it's not normal to do it this way but I did it because it makes everything more intelligible.


Here the results:


Luxcore Bidir on Ryzen 3990x

Basic set-up and advanced set-up - 8 minutes
View 01 1920x1080.jpg


Luxcore GPU Nvidia 3090

Advanced set-up - 2 minutes (is not the final one)
Lux_01.jpg
Light emitting objects had to be replaced by blender lights.


Cycles GPU Nvidia 3090 -

Advanced set-up - 2 minutes and 15 seconds
Cycles.jpg
Light emitting objects had to be replaced by blender lights.


Corona 7 on Ryzen 3990x

Basic set-up - 11 minutes
ryzen 2700x - 77 minutes .jpg

Advanced set-up - 16 minutes
ryzen 2700x - 107 minutes .jpg


Vray RT on Nvidia 3090

Basic set-up - 11 minutes
2060-70.jpg

Advanced set-up - 16 minutes
2060-100.jpg
Light emitting objects had to be replaced by Vray lights.


Maxwell Render 5.2 on Nvidia 3090

Basic set-up and advanced set-up 26 minutes
240 min 1060.JPG


Lumion on Nvidia 3090

Advanced set-up - 10 seconds
second render (3).jpg
Light emitting objects had to be replaced by Lumion lights.


White balance applied to all images in post.


Following are the settings to replicate the tests:

Camera position: 5,0,1.2
Camera target: towards the opposite wall
Camera angle
plan.jpg

Material and lighting description on the OBJ's layers.
Lux bidir has 24 light bounces, 256 ray bounces.

blender file https://drive.google.com/file/d/1I_qg7p ... sp=sharing
fbx file https://drive.google.com/file/d/1eIYmd_ ... sp=sharing
Last edited by daros on Fri Sep 24, 2021 6:46 am, edited 14 times in total.

daros
Posts: 223
Joined: Thu Dec 12, 2019 3:25 pm
Location: inside human language
Contact:

Re: Luxcore vs rest of the world

Post by daros » Sun Sep 12, 2021 5:47 pm

i'm updating the scene with a more appealing version

kintuX
Posts: 720
Joined: Wed Jan 10, 2018 2:37 am

Re: Luxcore vs rest of the world

Post by kintuX » Mon Sep 13, 2021 8:16 am

Interesting, but...

What about other engines like Cycles, Radeon ProRender, RenderMan, Indigo...?
Plus, it would be good to have full specs of the original scene (lights, colors, rays/bounces, ...) or preferably, attached as a reference. Especially, since Corona's results look quite different.

daros
Posts: 223
Joined: Thu Dec 12, 2019 3:25 pm
Location: inside human language
Contact:

Re: Luxcore vs rest of the world

Post by daros » Mon Sep 13, 2021 8:52 am

Yes i agree, i will also add indications about light bounces and raybounces. First let me update the existing results with the new scene, then i will try to test the other engines.

daros
Posts: 223
Joined: Thu Dec 12, 2019 3:25 pm
Location: inside human language
Contact:

Re: Luxcore vs rest of the world

Post by daros » Tue Sep 21, 2021 8:56 pm

I updated all the images with the new scene. Many rendering engines are missing i know.

But so far it seems that Luxcore Bidir is the only existing rendering engine that can be used for design verification.
Those who design lighting fixtures, crystal objects, car lights, museums, workplaces and so on, i.e. all those who need to know exactly how the light will behave in relation to their design, with luxcore bidir they just have to set the scene with correct materials and lighting and press render; the result will show exactly how the light will behave in relation to the design. As a result they can improve the design much earlier than usual.
With all other rendering engines you have to know beforehand how the light should behave and tweak the engine to produce something that looks realistic, something only extremely advanced users are capable of doing well.
Luxcore Bidir is in my opinion the ideal rendering engine for designers, engineers and architects.

kintuX
Posts: 720
Joined: Wed Jan 10, 2018 2:37 am

Re: Luxcore vs rest of the world

Post by kintuX » Wed Sep 22, 2021 2:57 pm

Well, you could also try Ocean Light Simulator for reference.

daros
Posts: 223
Joined: Thu Dec 12, 2019 3:25 pm
Location: inside human language
Contact:

Re: Luxcore vs rest of the world

Post by daros » Wed Sep 22, 2021 3:36 pm

i tried, but i cannot downlaod the software.

Martini
Posts: 116
Joined: Fri Nov 23, 2018 11:36 am
Location: Australia

Re: Luxcore vs rest of the world

Post by Martini » Thu Sep 23, 2021 11:08 pm

daros wrote:
Fri Sep 10, 2021 12:39 pm
Hi, this post is intended to do some justice to Luxcore's bidirectional rendering engine.
Just be really careful that you do not bias your testing methodology to try to prove a pre-determined conclusion. In general, it would be optimal to try to be as fair and unbiased as possible, regardless of the outcome, and then let the results determine the conclusion.

For instance, what were your constraints? Are you selecting a result where render time is fixed, sample count is fixed, similar amount of noise is desired, setup time is fixed, ...?

And what makes a "better" outcome? Least setup time, least render time, least perceptible noise, highest simulation accuracy, ...?
daros wrote:
Fri Sep 10, 2021 12:39 pm
Light emitting objects had to be replaced by blender lights.
Why? Most path tracers are capable of tracing light from emission materials.
i9 9900K @ 5000MHz | 2x Gigabyte AORUS RTX 2080 Ti Extreme Waterforce WB 11G @ 2010MHz
Z390 Aorus Pro | 32GB Kingston HyperX Predator RGB @ 4300MHz

daros
Posts: 223
Joined: Thu Dec 12, 2019 3:25 pm
Location: inside human language
Contact:

Re: Luxcore vs rest of the world

Post by daros » Fri Sep 24, 2021 5:12 am

Hi Martini, you are probably right, I should have explained the test better.
Yes, the various rendering engines had to show the highest accuracy given a certain level of noise, which is hard to fix because some engines make more noise on causthics and others make more noise on glossy white surfaces and so on. But this is a reasonable result, I think.
Anyway, the file is available for download and it's very easy to set up so everyone can experiment and draw their own conclusions.

But I think this test is not very biased... it's the common benchmarks where rendering engine weaknesses are kept hidden that are biased. This test simply requires the rendering engine to be able to solve multiple problems simultaneously, which is a very common need in production.

As for the issue of light-emitting objects, maybe I got something wrong, but these light-emitting geometries are apparently not digested very well by GPU rendering engines. GPU engines are simply not able to render them properly, so we had to replace them. In the Vray test you can see this clearly.

User avatar
Dez!
Posts: 324
Joined: Sun Apr 08, 2018 1:09 am
Location: Ekaterinburg
Contact:

Re: Luxcore vs rest of the world

Post by Dez! » Tue Sep 28, 2021 1:39 am

great work. a lot of conclusions everyone can draw for themselves and their workflow
Linux Plasma | Ryzen 5, 32Gb, SSD M2, GT 590 RX | BenQ 27 | Wacom One | Microsoft Ergo | Tie Guan Yin tea
http://dezigner.tilda.ws/

Post Reply