Hello,
For some reason I have this issue where when I raise the max secondary samples the image actually gets noisier. Is this expected?
This only happens with higher primary samples, like 250. If I set the primary samples to 10 or some low number, then having lower max secondary samples results in a noisier image and having more results in a cleaner image (as expected). At some point as I raise the primary samples this reverses. At 250 primary samples, if I set max secondary samples to 1 I get a cleaner image (and faster render) than if I set it to 9. I included the hip file.
What is going on?
thanks

update: seems to be the case where if raising the max secondary samples makes the image noisier, if I lower the variance threshold/noise level and render it again, it resolves a cleaner image. Still odd that raising max secondary samples would make the image noisier.

evanrudefx
update: seems to be the case where if raising the max secondary samples makes the image noisier, if I lower the variance threshold/noise level and render it again, it resolves a cleaner image. Still odd that raising max secondary samples would make the image noisier.

Probably a case of the two variances working against each other, raising one limit causes the other one to do less sampling but resulting in more perceptual noise.

Do you still see this when you use uniform sampling?

jsmack

evanrudefx
update: seems to be the case where if raising the max secondary samples makes the image noisier, if I lower the variance threshold/noise level and render it again, it resolves a cleaner image. Still odd that raising max secondary samples would make the image noisier.

Probably a case of the two variances working against each other, raising one limit causes the other one to do less sampling but resulting in more perceptual noise.

Do you still see this when you use uniform sampling?

That was one of the first things I tried. Even with pixel oracle set to uniform I still get noisier images with more samples.

the docs say to have noise level and variance threshold set to the same amount. Even if the variances were working against eachother, aren't the primary samples multiplied (or something like that) with the secondary samples. So I see no way in which increasing max secondary samples could cause lower quality render.
in other render engines, I have only ever had the issue of more samples taking longer (but look better), not more samples giving more noise and taking longer.