Passing through the null material counts as a bounce

[ameuret]

Hi buddies,

I just banged my head on the wall before realizing that crossing null material transitions counted in the depth (bounces) limit. I f I am not being clear, here are two pictures showing it off:

Depth limit set to 10 bounces

Depth limit set to 16 bounces

Forgive me if this is stupid but should a transition between air and air through the null material consume a bounce ?

This is using 0.7RC2 on Win7-64 with the path integrator but it holds true for the others too.

(the business cards are modeled as plain cubes, using a mix material with null for the cutout. Not perfect but very convienient since they are seen from a distance in the actual scene)

[jeanphi]

Hi,

Yes, this is a new behaviour since v0.7RC2, it is mandatory to fix some nasty bugs using null materials.

Jeanphi

[ameuret]

Arf, too bad !

A stack of 34 cards requires around 40 bounces, this slows down my scene by a factor of 2.

Thanks for the quick answer.

[Dade]

Arf, too bad !

A stack of 34 cards requires around 40 bounces, this slows down my scene by a factor of 2.

Thanks for the quick answer.

You should be able to tune russian roulette parameters and run about at the same speed than before

[Lord Crc]

If he's using LD and metropolis samplers for example, they will generate the samples needed for the full 40 bounces regardless of russian roulette parameters. This can indeed slow things down a lot in my experience, especially with bidir at higher depths.

[ameuret]

I did manage to get most of the original speed by fiddling with the RR settings, without losing convergence speed as far as I could tell. Here's the complete scene, nearly resolved (3h / 1800 S/px, Core 2 Quad 2.5 GHz).

Part of a business card design brief

[jeanphi]

If he's using LD and metropolis samplers for example, they will generate the samples needed for the full 40 bounces regardless of russian roulette parameters. This can indeed slow things down a lot in my experience, especially with bidir at higher depths.

Metropolis generates samples lazily so the speed penalty should be barely noticeable.
Maybe I could introduce the number of passthrough bounces as a parameter so that it doesn't interfere with the path lengths that much.

Jeanphi

[Lord Crc]

Ah, indeed it does, my bad... I've just noticed drops in speed with increasing depth when using metropolis over random.

Don't code and study for exams kids, it'll kill ya