Bidir vs Path

[gecko]

I realize there are fundamental differences in how bidir and path (and thus CPU vs GPU) process a scene, and that convergence on a solution will require different amounts of time depending on how the scene is lit (compounded by the fact that GPU is faster when the scene is optimized for it)... but I always assumed that ultimately the images should converge on roughly the same solution given enough time (ignoring any global illumination issues).

So here are two "simple" images rendered for 12+ hrs with the only difference that one used bidir and the other used path. Only lighting in the scene is a single sun/sky lamp. The syringes consist of a few different glass materials with volume priority correctly set, and tiny air gaps between all objects anyway. I realize that the existence of layers of different volume materials perhaps violates the definition of a simple lighting setup, but is this really the expected output of a path render? If so, I would suggest some additional restrictions on recommending path rendering "for scenes with simple lighting". Or did I just not understand how to set up such a scene for a path render?

Here's the bidir render, which is what I'd expect the scene to look like:

And here's what it looks like with path:

I tried all sorts of adjustments to get the path render to look "correct" (overlapping volumes with volume priority, completely air gapped volumes, turning subsurf modifiers on/off, etc), but it always looks like the render above. I typically use bidir for all my scenes anyway, but was excited about the idea of speeding things up by throwing my GPUs at it... Is there a fundamental reason this won't ever work?

[Egert_Kanep]

Maybe try to increase specular depth for path rendering, set it to 8 or more and see what happens

[Dade]

Maybe try to increase specular depth for path rendering, set it to 8 or more and see what happens

This, BiDir has a default of about 16 (8 eye path + 8 light path). Try to increase total and diffuse/glossy/specular max. depth to 16 in path too.

[B.Y.O.B.]

If the liquid directly touches the glass, you should not use air gaps. The meshes should overlap and the volume priority will take care to remove that overlap and create a correct glass-liquid border with correct IOR.

[gecko]

Maybe try to increase specular depth for path rendering, set it to 8 or more and see what happens

Wha! Thank you That was definitely the problem. So just out of curiosity, why does Path default to lower path depth than BiDir? I would naively assume the opposite - since Path is handicapped by only shooting rays from the camera, shouldn't it use more rays by default to make up for that?

If the liquid directly touches the glass, you should not use air gaps. The meshes should overlap and the volume priority will take care to remove that overlap and create a correct glass-liquid border with correct IOR.

Thanks for confirming, I thought that was the current best approach. Was trying other options in case the overlapping volumes weren't working right (since I know that the air gap trick "works" even if it's not quite right). It's also a difficult habit to break since I've done it for so long...

[B.Y.O.B.]

So just out of curiosity, why does Path default to lower path depth than BiDir?

Because longer paths = more noise.
And path is intended for more simple scenes than bidir anyway, so long paths should not be needed in most cases.