Cycles vs. Octane vs. anything comparison with images 2020

This thread is to compare rendered image times and quality of any render engines. This is not a place for opinion or arguments. If you want to prove render engine x is better or faster post images comparing the two with the same scene to prove the speed and quality. If something seems off in the images posted from what it should be share knowledge and scenes to try to get the images as alike as possible. Please also make sure to have all denoisers off. They make it harder to see grain and compare the real quality of the raw output of the render engines.

Some tips to match scene in different render engines.

  1. Lighting makes a huge difference in render time. This makes HDRI or area lights the most ideal way to match the lighting exactly.
  2. Colors of materials can render as different colors. When matching images the colors might have to be changed to get them to match right. For example the exact RGB of 120,156,186 might need to be something else for a different render engine.
  3. If the render engine has any filter try to have them all turned off. The filters could be the cause of the colors looking different. Here we are interested in stuff being the same not it looking pretty with filters.

It’s super hard to get render engines to be exactly the same. Good skill to all.

I’ll kick things off by copy pasting some tests I did from another post.
Done on 1x RTX 2070 super

Here is a link to the scene download. lacilaci86 is the one that directed me to it.
Here is the way to connect octane nodes for hdri if someone is having trouble. Make sure to have gamma at 1 not 2.2 like is default.

Room samples and bounces are same in both pictures.
Cycles RTX on, not eCycles

Octane RTX on
octane hdri1

I don’t have eCycles to compare, but eCycles should be able to do a more than 2x speedup depending on settings.

Same scene different hdri. Link to hdri
Same bounces and samples. Gray values of materials on scene are exact same value as well.

Cycles RTX on

Octane RTX on

Prorender, 1x RTX 2070 super

Scene 2
Link to scene and comparisons.

Edit: updated HDRI setup picture so it is right.
Edit2: added prorender for hdri2

1 Like

Where’s the link to the HDRI from the first scene?

Hi, I use second HDRI, something strange happen.
All settings default.
I have only a GTX 760 and I am a bit curious about the render times.

Blender 2.90 Cycles:
04:01.00 Minutes

Blender 2.90 E-Cycles:
02:05.00 Minutes

Octane 20.6:
02.29.00 Minutes
I have to set the power of texture to 4 to get nearly same lightning (Gamma 1).

Cheers, mib
EDIT: I forgot to disable CPU so I got 25% better performance than pure GPU render.
Opensuse Tumbleweed x86_64
Intel i5 3570K 16 GB RAM
GTX 760 4 GB /Display card
Driver 440.82

@Andrew_Ray that HDRI is a copyrighted HDRI which I don’t have permission to share. From testing more HDRIs than is shown here it seems the ratio of render times are the same regardless of the HDRI used. You can try your own or use the second HDRI or any HDRI.

@mib2berlin TY for testing E-Cycles. By all settings default I’m guessing the bounces were not set the same as I had to lower the bounces in Cycles. Default in Octane I think is 8 and Cycles is 12. Your numbers seem about right though. What sample # did you do? If you post images we could see if they look similar enough. Updating the numbers CPU disabled would be better comparison if you could run it again.

Hi, with default I meant open the file add HDRI and hit F12.
Checked again and bounces are set to 0 in your file, therefor my fast render times.
E-Cycles change tile size and order.
I can do more tests tomorrow, getting late here.
May you change your file with including the second HDRI, bounce settings and fix tile size.
32 tiles are really to small for GPU.
Make it easy for testers, open file, hit render.
I post a benchmark file some time ago and many user change anything to get better render time but this was very bad for comparison.

Cheers, mib

The room scene is not mine. My scene is the grass. The grass has the settings there all set up right.
For the room it’s not hard. Change the bounces to be the same and insert HDRI. Make sure HDRI is rotated same as Octane. Set same gray value on all objects. For your 4x texture power maybe you accidentally clicked on the jpeg instead of the exr in octane or didn’t rotate the hdri to the right angle?
It’s hard to know without any pictures posted.

This place is also for people to learn how to compare different render engines right. Posting a scene when people only press render would defeat the learning.

I did some test with the second scene using E-Cycles.
4096 samples, no AI de-noise — 01:28:07

512 samples with AI de-noise — 00:26:96

I know this thread is about Octane but now I’m curious. if I have time I’ll do a test with Redshift as well.

Edit: forgot to mention the specs.
Windows 10, AMD Threadripper 1950X, 64GB ram, Dual GeForce GTX 1080 Ti (Nvidia Drivers 441.28)

1 Like

This is a comparison between any any all render engines. I look forward and am excited to see the results from Redshift.

New render:
GTX 760
Bounces 6
512 Samples
Tiles 64

Cycles:

E-cycles:

E-Cycles settings changes but still 6 bounces:

Octane: With settings from @eklein except Power set to one.

It seams E-cycles give less noise with same sample rate.

Cheers, mib

1 Like

Here’s my renders with LuxCore on an i7-4790k and 2x GTX 970.

I think the lighting in the bottom of the bookshelf looks really good in LuxCore, due to the Photon GI Cache
30s render with Intel OIDN

120s render again using Intel OIDN

120s Raw

1 Like

Hi, Luxcore is really good and fast since last development, is this the Cuda core?

E-Cycles on Steroids (OIDN 1.2) and on low end GTX 760

46 Seconds:

watercycles,

In this scene to improve Octane performance by 30% from the defaults settings with no difference in image quality. Set the “Coherent ratio” to 0.35 and turn off “Alpha Shadows”.

It doesn’t make sense to use denoiser in the image comparisons. The point is to compare the path tracer performance and GI quality and use the same render samples.

Using a denoiser defeats the purpose of comparing render engines instead of denoisers. Octane denoiser is very good and could also get great image in seconds.

Here is the Octane optimize scene (no HDRI) with the correct HDRI rotation nodes. Download HDRI and render.

Octane Scene: Interior Corona No HDRI_octane.blend (4.4 MB)
HDRI Link: HDRI.

Hi, thanks for the scene, transform is not intuitive for me in Octane, I struggle with it since version 1.
You are right, denoising make no sense.
I edit my Octane render in my post.

Cheers, mib

Win10, GTX1080. Aging i7 [email protected] No denoiser.

eCycles GPU 4096 samples, 6 bounces.

ProRender GPU 4096 samples.

@mib2berlin your hdri seems to be mirrored different in the Octane render than all the rest. Make sure the Scale is not negative for the HDRI.

@zeealpal LuxCore looks impressively fast. I’ll have to try that out.

@eklein Thank you for reinforcing denoise should be off. I stated this in the first post. Disabling Alpha shadows in Octane is a good idea for this scene. Upping the Coherent ratio adds flickering when animating so I’m going to say that is a no go. Any method used must be flicker free in an animation. With as fast flicker free rendering can be done now a days it’s a must even if it takes longer to render.

Here is the results of Alpha shadows off. 10 sec. faster then with Alpha shadows off.

Luxcore has the prettiest GI :heart:

Thanks, I am using a GI cache, so if it’s an animated scene (excluding camera) I will test to see if there is any flickering.

If only the camera is animated the same cache can be reused between frames.

i think this is very unfair comparations (like all other topics on the forum - where simple cubes are compared, or geometry with a gray shader.) - because this scene is very simple with very simple lighting and simple shading:
render engines like Octane, Vray (vray Proxy, vrayScater, lightCache, outOfCOre) fstorm (memory paking, veryfastGPUrendering) of Corona (best and fastest use CPU for render) start shine when there are scenes with many threes (or other complex geometry)













Or interior with complex shaders and can compare quality of GI base on how shaders itself interoperate between themselves and it self make GI and atmosphere between themselves base on secondary rays of shaders.






or scenes where so many geometry that not fit in VRAM and octane cant render scene but fstorm CAN just because it have some exotic memory packing for geometry…






cycles is good (and i love it), very good for it situation but other engines is have dozens of developers hwo throu their time only on render engine and matematics and tehnologis behind corona render or fstorm render simply not exist in cycles and when some one who spended many time woth corona render use cycles - the lack of technology constantly appears here and there, as soon as you do something more or less complicated. Yes, you can always get around something somehow - but this is done only because these technologies simply do not exist.

if anyone wants a fair comparison of rendering engines, then the scene should be complex, affecting all aspects of the engine, loading a lot of memory, containing all the main components (SSS, multyLights, proxy system of unload geo, GI, complex shaders) of the engine…

Again i do not say that cycles cant do all this things - it can, but very limited ( with all sorts of tricks) and slowеr than others engines (and comparison of simples scenes in general have no sense)


1 Like

Well, provide us with a complex scene then. There is a reason why most test scenes are relatively simple, because it is hard to create a complex scene that is compatible with all renderers and that can serve as a good comparison. At least a scene such as this one provides a baseline. And not everyone renders highly complex scenes - rather the opposite.

Anyway, why are you posting 20 corona rendered images here? Why not use Corona to render that same baseline scene first? Or was it too slow for your taste to include here?

You seem to love Corona, so could you render this scene in Corona, and post the result here? I think it would be interesting to compare a CPU-based render engine as well.

because it is hard to create a complex scene that is compatible.

this does not mean that the comparison simpliest gray shaded scenes is correct and valuable

why are you posting 20 corona rendered images here?

Pictures are examples of my words. Do you forbid me to show pictures to confirm my words?

Or was it too slow for your taste to include here?

lol - who care about time with simply gray shader? BUT! when you shade complex scene and when exit from local view scene simply not fit in 11gb vram - that is sad. Or when complex scene more faster render on CPU than gpu, or when complex scene in cycles after 5000-10000samples can’t get reed noise from dark areas?
After that-
Here then nobody is interested in simple scenes in simple gray shader.

so could you render this scene in Corona, and post the result here?

I have no desire to prove something to those for whom simple scenes in simple gray shader something mean.

For myself, every day, over the past many years, I confirm what I am writing here.
And all that I wanted to say and show - I wrote in my message. To prove to someone that when you see for decades what you are talking about in practice, there is no desire to prove anything to anyone.

All is said.

1 Like