New Mental Ray shaders development

[Kzin]

did some tests on weekend with mila.
its a wip model with some 2d shaders for glossiness and anisotropy. i used alot of hdr reflection planes for the upper lights.
for lighting i used 12 lights, all user_ibl_rect arealights with self created hdr images of lightsources which lit the scene. fg with 2 bounces in addition.

the simple wall image took 68 minutes and the image with reflective walls took 3h45min but i am trying to reduce it. raydepht is 3 for all glossis.

[Kzin]

i rerendered the last one but this time with local sample set to zero, like for all other glossi surfaces.

rendertime is now 2h8min compared to 3h45min. its alot faster, so better is to use zero local samples (internally zero means 4 samples) first.

unified settings was min1/max300 with quality of 2 for all renders. the good thing about the new shaders is that you dont need high quality values anymore.

[ActionArt]

Interesting. For a render time that long it seems very grainy, especially the shading in the top right corner and the shadows off the supports. I wonder what it would take to clean that up.

[Kzin]

Interesting. For a render time that long it seems very grainy, especially the shading in the top right corner and the shadows off the supports. I wonder what it would take to clean that up.

which grain do you mean? there is only a small amount of grain in the image.

[ActionArt]

Mostly quite noticeable in the top right corner on the wall of the first image. From FG maybe?

[Kzin]

Mostly quite noticeable in the top right corner on the wall of the first image. From FG maybe?

ah, ok, that will be not enough arelightsamples of the 2 lights from top, should not impact the rendertime that much. the importance sampled ibl_rect shader renders very fast and i am using not that much samples.

as you can see in the second image its not visible anymore, so i did not changed it. i thought you mean the noise like reflection walls, but its only a bumpmap.

and i dont think it renders slow with all that lighting and shading stuff in the scene.
but i will add some stuff from time to time to make it more complex, more objects, so more shadows and more reflection complexity.

[ActionArt]

Ok, that would explain it.

Yes, the render time is not too bad overall. It would be nice to somehow compare it somehow with the current shaders and just FG, no ibl.

[Kzin]

Ok, that would explain it.

Yes, the render time is not too bad overall. It would be nice to somehow compare it somehow with the current shaders and just FG, no ibl.

i am using fg with two bounces. its a closed room. the 12 area lights are using hdrmaps to lit the scene. but i used the ibl_rect light shader for the lights because the light is using importance sampling which makes the whole thing alot faster. the lights are rectangular arealights with the ibl_rect shader and hdrmaps as inputs.

so its not lit by an spherical hdr envmap like the normal user ibl.

[ActionArt]

Interesting. Any idea how much time you save using those lights over standard rectangular area lights or regular rectangles with hdr maps attached?

[Kzin]

Interesting. Any idea how much time you save using those lights over standard rectangular area lights or regular rectangles with hdr maps attached?

tested it with another model and with physical lights it rendered about 2 to 3 times slower (the setup in image one without the glossi walls).

[Kzin]

did another test, rendertime is 5 hours but i have a problem for this render.

' INFO : RC 0.4 1396 MB info : type number per eye ray
' INFO : RC 0.4 1396 MB info : eye rays 104501478 1.00
' INFO : RC 0.4 1396 MB info : reflection rays 399113066 3.82
' INFO : RC 0.4 1396 MB info : shadow rays 7151156029 68.43
' INFO : RC 0.4 1396 MB info : probe rays 1272974511 12.18
' INFO : RC 0.4 1396 MB info : fg points interpolated 501536537 4.80


as you can see it has rendered over 12 probe rays per eye ray but i dont use any ao at all. so i have to investigate what this probe rays caused. i hope its not a bug that calls ao rays also when its deactivated, will do more tests later.

[Kzin]

found the problem with probe rays, its the bump map which causes it. of course, rendertime without bump is better, so i cant say how much impact the probe rays have. at the moment its also unclear if this behavior is correct, will let it check.

in the meanwhile, another rendering with max 100 samples for unified. a bit noise in the glossis, but i think it would be ok for animations, for stills i would raise max samples to 200 again.
i should also say that i changed the shader a bit including the last render. i added a second reflection layer to get better indirect reflections like you can see. and i dont use any reflection falloff, so what you see is brute force in a room which is over 20m in high, about 60m long and over 20m in width. so qualitywise its really good. rendertime could always be better, no doubt. ;)
this one took 4h and 8min.

[Kzin]

new rendertest, this time with (coarse) grass. again 2 layer for glossis in addition to a diffuse_reflection and 1 mila_scatter node for sss.

rendertime is about 6h30min.
a bit long at the first view and i am sure i can save same area light samples. but for the shadingcomplexity its not that bad.
its only 1,5 hours more then a test i did with the engine in the grass lit by an user_ibl. and this is inside a big scale room with 2 bounces fg, 12 arealights and 3 raydepth glossis everywhere.
why i am that positive is that the rendertime does not explode that much when in earlier testscenes.
i run alot of tests with mr and always had the problem of rayexplosion, rendertimes of 7 or 8 hours for some things in smaller res let me go crazy. but mila shaders are working waaaay better.

[Kzin]

ok, new update but it has changed alot because there were something wrong, or lets say not good, setup wise.
first the diffuse, i set the diffuse fresnel to most mats wrong. it was from the basic and fast setup to test the shaders. the fresnel is now set to default for all shaders.
and second the indirect and direct reflection nodes was not weighted correctly for most mats. i also changed this, so indirect has now the same weight as direct and vice versa.
third the lights, i decreased the arelight samples for the lights from the side and also added 14 more of them, now we have 26 arealights in the scene.
and i added another wip object which is partly textured. 4 to 6k textures with dispmap enabled. the glas contains of 1 diffuse transmission node and 1 glossi reflection node. booth have slightly glossi values.
rendertime is about 5h30min with little noise in glossis.



Moderator edit: as this - understandably - is turning into a very "image heavy" thread, you might want to consider switching to our rimg= tags - HB

[Kzin]

did a small carpaint shader, it has 1 diffuse, 4 glossy reflection and 2 different bump flake layers.

[Rez007]

I haven't touched the new shaders yet, but the layering system is looking really cool. Thanks for posting the images.