OTOY Forums

Hello, guys! But this thread is very important!

Render caustics are (almost) cleaned:

Sun sizes / resolutions screen - 1002x602 .... / 1920x1080 . / 3840x2160
0.1x (on Saturn)........................... - 13,500k spp* / 4,000k spp* / 1,000k spp*
1x (on Earth)................................ - 260k** .......... / 75.5k ........... / 18.9k
2x.................................................. - 78k ................ / 23k ............... / 5.75k
4x ................................................. - 23k ................ / 6.8k .............. / 1.7k
8x ................................................. - 7k .................. / 2.1k .............. / 525
10x ............................................... - 4.9k ............... / 1.45k ............ / 360

"spp" - samples per pixel

VERY IMPORTANT: do not forget to disable "caustic blur" from default at PMC settings. If we set to 10k-reject to may get some noises (noises colored if any specular material has dispersion) then we try more spp.

* - we can't increase "Max. samples" from 256k using OR 2.24.2 But IMPORTANT!!! - we'll change "Max. samples" from default as 500 to 0 to get infinite spp using newer version of OR.

Do you like this thread?

If you want any other planet then contact me.

I investigated caustics... but I would need to increase from 23 kS/px to 50 kS/px at sun size 4x because this increase of S/px reduces noises colored for other things (not caustics).

Sun size 4x at 25 kS/px

This image has to seem few noises colored.

Sun size 1x at 256 kS/px

This image is cleaned and looks pretty well!

PMC settings:
27-depth both diff+spec
10k-reject
parallelism 1

"Ice" material:
reflection 0.930
transmission 1.000
ior 1.299
coefficient dispersion 0.003


--> Do you like that?

ice in the material necessary to tick the fake shadows

Elvissuperstar007 wrote:ice in the material necessary to tick the fake shadows

I do never like fake shadows as fake caustics because it looks BAD caustics rendered = no "real-life" and I deserve much "real-life". lool

But I wait new kernel as maybe progressive photon mapping (PPM) for much shorter time render than PMC.

nuno1980 wrote:

Elvissuperstar007 wrote:ice in the material necessary to tick the fake shadows

I do never like fake shadows as fake caustics because it looks BAD caustics rendered = no "real-life" and I deserve much "real-life". lool

But I wait new kernel as maybe progressive photon mapping (PPM) for much shorter time render than PMC.

Whats lool? "Laughing out out loud"...

But anyway, you do a very good job by giving unnecassery information to the community

djart wrote:Whats lool? "Laughing out out loud"...

yeah. But "lol" - looool, lolol or lloll...

See differences of caustics between 3 images at S/px and resolution screen about teaport:

- at 360x216 - 150 kS/px


- at 1002x602 - ~19,337 S/px

360x216 and 1002x602 seem to be same caustics and same noises.

- at same resolution screen - 150 kS/px


Higher resolution screen = better caustics and less noises at 150 kS/px but longer render time.

I'll buy OR in my anniversary (on july) but maybe few next months.

--> Do you like this? - but IMPORTANT!! Should respond me it.

Good tests nuno, have you tried by rendering at double or quad for few samples and then down scaling at the original resolution?
ciao beppe

teapot test is not ok,
when you involve many objects in interior spaces you will need 8 Titan X to get a clean image.
Fox exteriors PMC is useless, pt is a lot faster especially if you set GI clamp to 1. Also using PT with GI set to 1 you have to use caustic blur if you have metallic or very reflective objects in the scene.
For interiors you can use pmc but you need 8 fast gpus, it's better to use pt, Octane will generate sharper images, well defined texture and less noise in dark areas.
I never used glossy and specular samples at high values except you have several glass layers, I always use 8-16, rarely 24.
Caustics are useful only if you create a pool.
Don't use dispersion because it generates noise.

regards

bepeg4d wrote:Good tests nuno, have you tried by rendering at double or quad for few samples and then down scaling at the original resolution?
ciao beppe

No, 3 images are original. But I'll downscale to 1280x800 (eg) for blur by AA noises from 2560x1600 (eg) after OR bought.

For downscale, open any PNG image with Paint, zoom out from 100% to 33% or 50% to lower resolution screen and capture (press keys ctrl+print) it. Run 2nd Paint program to paste this image captured and cut from it to image rendered.

gabrielefx wrote:teapot test is not ok,
when you involve many objects in interior spaces you will need 8 Titan X to get a clean image.
Fox exteriors PMC is useless, pt is a lot faster especially if you set GI clamp to 1. Also using PT with GI set to 1 you have to use caustic blur if you have metallic or very reflective objects in the scene.
For interiors you can use pmc but you need 8 fast gpus, it's better to use pt, Octane will generate sharper images, well defined texture and less noise in dark areas.
I never used glossy and specular samples at high values except you have several glass layers, I always use 8-16, rarely 24.
Caustics are useful only if you create a pool.
Don't use dispersion because it generates noise.

regards

My test is good!!

Your tips about 8 GPU's is good but I haven't very big ("professional") money. But I'll upgrade to new GeForce x80 Titan (Pascal) in Christmas and Pascal with HBM2 will be ~10x faster than Maxwell at GPU computing only. After X80 Titan released, we'll review surprise and very good result using OR.

Do you deserve "real-life"? If you respond me "yes" then should correct render and specular material settings:
- PMC kernel with:
24-depth (diff+glossy) for 1 clear glass only or between 100- and 300-depth for many clear glasses and/or few mirrors or for 1 3D star glass;
paralellism 1;
no caustic blur;
max GI;
5k or 10k reject.

- specular material with:
reflection between 0.9 and 0.98;
correct index of refraction* if it has dipsersion then should reduce índex dispersed as índex original from index original;
must enable dispersion with correct value*

* - see https://refractiveindex.info/ and do formulas**

** - formulas:

Code: Select all

B = (n_d - 1)/V_d*((0.48613^2*0.65627^2)/(0.65627^2 - 0.48613^2))

F and C = F and C wavelengthes - If wavelength = 589.3 nm then should use F = 486.13 nm and C = 656.27 nm


n_d_dispersed = n_d_original - 10/3*B

B -> dispersion coefficient
n_d -> refractive index
V_d -> abbe number
n_d_dispersed - refractive index dispersed if the dispersion is enabled, for Octane Render software only

If you have any doubt about formulas, contact me.

You may not learn abbe number/dispersion because you haven't Chemical and/or Laboratory required.