Hybrid Back/Forward path tracing (aka BiDir without MIS)

[B.Y.O.B.]

Not sure if I understood everything correctly, but the OpenCL settings marked with red dots in your last post only affect the imagepipeline (they are basically film.opencl.enable and film.opencl.device).

[epilectrolytics]

I agree, it is very clear:

As B.Y.O.B. mentioned, your markings show the device used for the image pipeline, not renderer, as specified in Blender.
The other image from my GIF shows the render settings with the changes I made in the file:

It is not possible to achieve a render result like that, showing caustics only, without switching to PathCPU first and then setting light tracing to 100%.

I'm sorry that my post is causing confusion and will post a config file later when I'm back at my PC.
That will have the settings included and show the weird render out of the box.

[Dade]

I can't seem to get denoising to play nicely. Sobol with OIDN works, but BCD doesn't show any light tracing. When using random, any denoising causes light tracing to be disabled entirely.

Yes, I have probably forgot to add the BCD support to Hybrid. But, sincerely, I don't even know if it is worth doing, I will check if it can be done without too much effort.

I should have fixed this problem.

[Dade]

Seems there is still a problem with motion blur.
I tried to render the classroom scene with hybrid path, 100% light tracing for testing and encountered a strange result:
- caustics seem way too bright
- upper half shows a strange, different pattern
- with motion blur disabled the render seems ok

It wasn't a problem related to Hybrid but to light tracing in general and it was affecting BIDIRCPU and LIGHTCPU too. I should have fixed the problem.

[epilectrolytics]

It wasn't a problem related to Hybrid but to light tracing in general and it was affecting BIDIRCPU and LIGHTCPU too. I should have fixed the problem.

Thanks, now everything works fine again!

will post a config file later when I'm back at my PC.

I think that's obsolete now

[lighting_freak]

Hi,

I have a scene that represents a common lighting scenario.
A row of LED illuminates a volume diffuse, but mostly polished, lens.
All this is built in a housing and closed by a clear cover lens.

This is the scene:

HYBRID_FAIL.blend

(2.43 MiB) Downloaded 214 times

This is the result in classic CPU BiDir:

This is the result using hybrid (several different settings - all the same result image):

Do you believe I'm doing something wrong, or is CPU only not dead yet?

BR

[provisory]

Do you believe I'm doing something wrong, or is CPU only not dead yet?

I think classic BiDir is definitely not dead yet.

Hybrid PT is the most exciting development for me in LuxCore, but there are still problems with it.

- It's a known limitation, that it can't do SDS paths. Dade says, that BiDir can't do either, but with BiDir + metropolis + a lot of patience you can get something, however with hybrid it's just impossible.

- It seems to me, that hybrid doesn't handle glossy reflections correctly, I mean in volumetric light a metallic reflection with roughness 0.2 looks like a total specular reflection.

- Volumeric lights often appear much brighter with hybrid than with BiDir.

I'm away from my pc for a couple of days, so I can't upload test scenes now.
Furthermore I understand, that Dade has more important tasks now to finalize 2.2 release.

[lacilaci]

yes yes, sds "fix" needs to happen before killing cpu only rendering.

Dade already mentioned that caustic cache can be modified to render only sds.

But I guess that will happen after 2.2 or maybe even after 2.3?

2.2 still needs enviro light caching fix and support for opencl and 2.3 is supposed to be about materials.

But I do wonder how the sds only caching for caustics will work. will the entries be created by lighttracing only? And then have the opencl use cache entries for sds only? I'm a very simple person so this sounds super complex to me.

Pathtracing(gpu) lighttracing(cpu) gicache(gpu) lighttraceSDScache(cpu) and somehow magically merge and have everything play well with each other?

[Dade]

But I do wonder how the sds only caching for caustics will work. will the entries be created by lighttracing only? And then have the opencl use cache entries for sds only?

Yes, the cache will be created a pre-processing time like current caustic cache and it will be used only to render the result of SDS paths.

Pathtracing(gpu) lighttracing(cpu) gicache(gpu) lighttraceSDScache(cpu) and somehow magically merge and have everything play well with each other?

Yes, they all solve different parts of the final result (i.e. there is no overlap) so they can be just added together.

[Sharlybg]

Light tracing doesn't work with Camera clipping plan :

ERRR.blend

(784.05 KiB) Downloaded 216 times