Help With Speeding Up Render

[amm]

3 hours ago, Anti-Distinctlyminty said:

So, I suppose the question remains that should this be considered a bug? Is the transparency noise caused by errors in roughness calculations or are they inherent to PBR? I would naively expect similar behaviour of noise for both reflection and refraction.

In case of glass, there's old trick utilized in some custom Mental Ray shaders, to distribute sampling, reflection *or* refraction, which on is more prominent for certain pixel. Worked well ten years ago, using machines of these times. Don't know what happens there, anyway result is really bad and unexpected. I've also tried your scene, tonight.

Edited January 4, 2018 by amm

[TobyGaines]

On 2018-01-04 at 4:10 AM, Anti-Distinctlyminty said:

So, I suppose the question remains that should this be considered a bug? Is the transparency noise caused by errors in roughness calculations or are they inherent to PBR? I would naively expect similar behaviour of noise for both reflection and refraction.

I was using Raytrace, not PBR, because that's what your scene had, but they render this the same - micropolygon looks even worse

I think it's the principle shader with GGX that's causing most of the problem - when I use classic shader core with Phong, there's no noise. This is one of the cards under your petri dish, no reflection

( I cranked up the gamma, on both, to show it better )

[davpe]

15 minutes ago, TobyGaines said:

I think it's the principle shader with GGX that's causing most of the problem

yeah that's what i've learned too, that classic shader performs significantly better in this case.

ggx is supposed to give more plausible rendering than phong (and it does), unfortunately it also means that insufficient sampling is more apparent. phong is showing no noise but also lacks some detail. 

[Anti-Distinctlyminty]

Just returning to this after a while, as I was gain having significant issues involving render times for transparent objects and had a few observations of light quality sampling that was somewhat confusing. I have an item in my scene that is using the aluminium preset, so it's using PBR metallic. 

Each gif shows two frames each of the direct lighting. The first has a light quality setting of 1, the latter 10. There's practically no difference (apologies for the slightly differing sizes)

 

Now, the thing is, when rendering a material that uses PBR Reflect, that's the complete opposite. Here's a render, again of the direct lighting, of the rim of an item, with a light quality setting of 1

And now, with the light quality of 10

 

 

 

So, I'm somewhat confused...does the light quality setting only have an effect on PBR Reflect, and not PBR Metallic? This poses several problems also...just to get this one specular hit on an object I have to massively oversample the light for everything else in the whole scene.

I also attached the scene if anyone wants to take a look at that.

 

SESI_Render_Issues.zip

Edited February 11, 2018 by Anti-Distinctlyminty

[Anti-Distinctlyminty]

Here's something interesting...I tried to rebuild a simple pbr compatible shader using generalfresnel, reflectedlight and refractedlight vops. The result is a lot cleaner than using pbrreflect. You can reconnect that to see the difference.

I've attached the file so you can take a look in /mat/Top_Clips

Weirdly, the direct light reflection from the light in the scene looks a lot more solid, however changing the quality setting on the light now has no effect whatsoever. So, if the result isn't quite as good as you need, you're out of luck and just have to increase pixel samples, again oversapling everything else in the shot.

SESI_Render_Issues.hip

[mestela]

That sounds like a nice repro to send to sidefx, good thing to poke Kai about.