BDPT vs PPT

[Rob Burns]

This might have been discussed but after searching and reading quite a bit about both path tracers I still have a few questions about them. I'm looking for a high-end rendering method for portfolio images, something that captures as many subtle, realistic lighting effects as possible, like accurate reflective caustics and others so I've been experimenting with the path tracers. Render time is not really an issue for this and I don't mind letting them just run for days if needed to get a clean result. A few questions:

Is there a difference in the lighting effects that PPT and BDPT can simulate. Assuming I set the sample size in PPT to 0 so it is unbiased, will they both eventually converge to identical solutions, or are there some differences in the lighting effects they can produce? In other words is one more realistic or accurate than the other?

Down the road will either of them become more developed than the other, or be more capable of simulating realistic lighting than the other?

Can either or both of them work with distributed rendering? I did some speed tests a while ago with the BDPT and it seemed in ran faster with distributed rendering, but I might be remembering it wrong - would be good to get an official word on this. I'm assuming PPT cannot be distributed since I don't think the Light Cache can be distributed and I think PPT is basically the same algorithm, but again would be good to get some input from the experts.

thanks for the great work here!

[vlado]

Is there a difference in the lighting effects that PPT and BDPT can simulate. Assuming I set the sample size in PPT to 0 so it is unbiased, will they both eventually converge to identical solutions, or are there some differences in the lighting effects they can produce? In other words is one more realistic or accurate than the other?

They both calculate the same lighting solution, however there are some effects that BDPT can calculate but which are not possible with PPT (i.e. caustics from point light sources or directional light sources). For the effects that both can calculate, they are both accurate, however one may be faster to converge than the other. The PPT algorithm is better tested, so if render time is not an issue, it might safer to go that way.

Down the road will either of them become more developed than the other, or be more capable of simulating realistic lighting than the other?

The bidirectional path tracer will get vertex merging, which will improve convergence for a further class of lighting effects (i.e. caustics from point lights or directional lights seen through a mirror).

Can either or both of them work with distributed rendering? I did some speed tests a while ago with the BDPT and it seemed in ran faster with distributed rendering, but I might be remembering it wrong - would be good to get an official word on this.

The bidirectional path tracer supports DR in the latest service pack; it also supports the Light select render elements.

I'm assuming PPT cannot be distributed since I don't think the Light Cache can be distributed and I think PPT is basically the same algorithm, but again would be good to get some input from the experts.

This is correct, yes. For the next major release of V-Ray (V-Ray 3.0) we also have a progressive image sampler, which works fine with DR.

Best regards,
Vlado

[Rob Burns]

They both calculate the same lighting solution, however there are some effects that BDPT can calculate but which are not possible with PPT (i.e. caustics from point light sources or directional light sources).

Thanks for the info Vlado! Would the VraySun be considered a directional light? Does that mean that neither path tracer can simulate caustics from the VraySun? In tests that I did I thought I was getting reflected caustics from objects illuminated by the sun but maybe it was something else. I am a bit confused if 'GI caustics' and just 'caustics' are the same thing or different, could you shed some light on that? Cheers.

[vlado]

Thanks for the info Vlado! Would the VraySun be considered a directional light? Does that mean that neither path tracer can simulate caustics from the VraySun?

They will generate caustics from the sun in some cases. Both engines will generate caustics caused by the sunlight being reflected in glossy materials. BDPT can also produce caustics caused by sunlight reflected off a perfect mirror surface.

In tests that I did I thought I was getting reflected caustics from objects illuminated by the sun but maybe it was something else. I am a bit confused if 'GI caustics' and just 'caustics' are the same thing or different, could you shed some light on that? Cheers.

Caustics are caused by light being reflected non-diffusely off a surface, or refracted through it. The effect is just one, but there are different ways to compute it.

Best regards,
Vlado

[Rob Burns]

Great info, thanks