'samples' parameter does not work on Skydome lights (Arnold-5)

[alexanderVLZLJ]

Is it normal that skydome light does not respect samples parameter in Arnold 5.0.1.3? I have checked the release notes and could not find any mention of this parameter becoming obsolete. What am I missing?

Thanks, Alex

[Stephen.Blair · Accepted solution]

There's no light sampling for constant skydome lights, just BRDF sampling. That's an optimization.

If you plug in an HDR, you'll see a difference in the number shadow rays (shadow rays are for light sampling).

samples = 1

| -----------------------------------------------------------------------------------
| ray counts:                      ( /pixel, /sample) (% total) (avg. hits) (max hits)
|  camera                  4799790 (   9.26,    1.00) ( 90.41%) (     0.01) (       1)
|  shadow                   263152 (   0.51,    0.05) (  4.96%) (     0.00) (       0)
|  diffuse_reflect          123069 (   0.24,    0.03) (  2.32%) (     0.00) (       0)
|  specular_reflect         123143 (   0.24,    0.03) (  2.32%) (     0.00) (       0)
|  total                   5309154 (  10.24,    1.11) (100.00%) (     0.01) (       1)

samples = 8

| -----------------------------------------------------------------------------------
| ray counts:                      ( /pixel, /sample) (% total) (avg. hits) (max hits)
|  camera                  4799790 (   9.26,    1.00) ( 75.32%) (     0.01) (       1)
|  shadow                  1326306 (   2.56,    0.28) ( 20.81%) (     0.00) (       1)
|  diffuse_reflect          123069 (   0.24,    0.03) (  1.93%) (     0.00) (       0)
|  specular_reflect         123143 (   0.24,    0.03) (  1.93%) (     0.00) (       0)
|  total                   6372308 (  12.29,    1.33) (100.00%) (     0.00) (       1)

[alexanderVLZLJ]

Thank you for the answer, Stephen. It would be great if changes like this were documented and added into release notes. This 'optimisation' made certain types of renders significantly slower and we were not sure what the cause was. We were relying on skydome lights for our animation renders (playblasts). A combination of high-AA samples (6x6 to resolve motion-blur) and low-light samples was giving us a good balance of quality vs render time. Now this option is not available anymore since skydome light will never use less than 4 samples. I do understand that using a single sample for a skydome light sounds ridiculous, but it was working just fine for playblasts if we kept max shadow-distance to a reasonable small value. Now all our playblasts are more expensive than they were in Arnold-4.

[Stephen.Blair]

I don't understand. You can still set the skydome light samples to 1

For constant lights, 1 is the effective maximum

[alexanderVLZLJ]

You are (partially) right, Stephen. I did some more tests and it turned out that the the extra shadow samples are coming from the shader not from the light. The new shader has specular layer ON by default, which accounts for one extra sample. However, even if you disable all layers (diffuse=0, spec=0, coat=0, etc) there are still shadow samples. Where are they coming from?

diffuse+specular (default): 3 samples per light

diffuse only: 2 samples per light

black: 1 sample per light

[Stephen.Blair]

If I set all the standard_surface weights to 0, I don't get any shadow rays. This is with one sphere and a skydome hdr.

[alexanderVLZLJ]

Stephen, it turned out that the extra samples were coming from shadow_matte aov that was enabled in my test scene. Area lights and skydomes with HDRIs do not seem to be affected by this aov, but uniform-color skydomes do. This is not a big deal, but it is a bit confusing.

Thank you for the follow-up, Alex