Quote from: blattacker on March 29, 2024, 02:13:19 PMDoing some quick experimentation on my end, the lambert shader does seem to take longer to render with the path tracer, but color seems to have a more pronounced effect. For reference, the experiment I did was just two simple shape shaders with no other displacement, and I tested rendering with the default "Base colours" pf shader, the default pf edited to introduce color rather than just greyscale values, a lambert shader with a grey value, and a lambert shader with a color value. All other setting remained the same. My results were:
- Render Settings: 800 x 450; 0.5 Micropoly detail; 3 Anti-aliasing (all default settings); Path Tracer; Max paths per sample 144 (high value to approximate a "worst case scenario")
- Greyscale PF shader: 54 seconds
- Color PF shader: 2 minutes 43 seconds
- Greyscale lambert shader: 2 minutes 30 seconds
- Color lambert shader: 8 minutes 1 second
Perhaps a workaround to try it out with the path tracer would be to render it out in greyscale and then either add color in post or process it like a black and white photograph?
Thanks for taking the time and effort to look into this. I already did similar tests last week and reported my findings to the team. Something might be going on, but may be also not. Let's see.
My conclusions are similar, somehow bright saturated colours take a big hit in this situation.
It's fine the PT is slower, but usually not 100x. Rather 1.5-10x slower.
Your findings are similar to mine, but some info is lacking.
For example, lambert and surface shader render at equal speed for me.
PF is faster, but that's because the low colour is black. Setting both to the same colour as lambert results in only slightly slower render, which makes sense since to me.
Your suggestion at the end can't work, because with grey colour you basically omit GI. All the various shades and saturations of colours you see in the renders come from GI.
Quote from: blattacker on March 30, 2024, 04:20:51 AMQuote from: Dune on March 30, 2024, 03:35:27 AMPerhaps the (expensive) way of calculating light?
I think that might be the case. If my (extremely basic) understanding of rendering principles is correct, I believe Lambert shading is a form of reflective shading, albeit diffused reflections. I would guess that those diffuse reflections require additional calculation. Moving into pure speculation, color would likely increase calculation time as it would reduce the amount of bulk operations that could be performed, since each ray could have different color values based on what color(s) it picked up (or, as it works in the real world, I guess it would be which colors/wavelengths got absorbed rather than reflected) along the way, but again, that's just pure conjecture on my part. I'm not quite clear on the scope of the physical aspect of physically based rendering.
Lambert shader is a diffuse shader. Nothing special going on with that compared to the colour from a PF or a surface layer. Those are lambertian models too.
My lambert also does not use translucency and the (test)scenes also don't have reflectivity. It's only 1 simple diffuse lambert shader.
As I said normally PT renders 1.5-10x slower, depending on a lot of scene-related factors, but this performance hit is unusual in my experience.