Page 6 of 6
Re: Cold Fusion
Posted: Sun Jun 16, 2019 9:44 pm
by Fox
@ kintuX
It's pretty old Xeon e5-2600 v4. The GTX 970 is around 6 times faster.
Re: Cold Fusion
Posted: Mon Jun 17, 2019 8:44 pm
by epilectrolytics
Fox wrote: ↑Sun Jun 16, 2019 8:33 pm
I got this scene to render pretty clean with OIDN under 12 hours, seems to even work with classic BiDir.
With multi scattering enabled and russian roulette disabled, i was enable to make 1 pass to take 2500 seconds (bidir 64/64 x BiDirVM 256)
Multi off and russian roulette on, i got nice clean render.
On that render the multi is on and the russian roulette value is set from default 3 to 6. I should go back to multi off and RR at 3. Even OIDN has problem giving good denoising.
Here with scattering asymmetry down from 0.9 to 0.5 the light beam is stronger, but with classic engine the beam is blown out inside prism, don't know if the spectral rainbow will come out, perhaps it will.
55 passes.
55 Passes, multi scattering on.jpg
Prism BiDirVM.7z
Thanks Fox for explanation and providing a blend file, that helped me a lot!
I learned that mixing dispersive and rough glass together with high scatter value makes the internal beams appear even in classic BiDir.
The magic you did with the noise driven asymmetry is beyond me but certainly crucial for the effect you get in your image.
At least now I got my own version working, rendered with BiDirVM (I come to understand slowly how this is irresistible) but was not patient enough to wait some hours because suddenly Dade introduced a cyclic caustic cache wich I needed to test
Some dust&scratch removal in PS:
Re: Cold Fusion
Posted: Wed Jun 19, 2019 9:00 am
by p2or
Fox wrote: ↑Sun Jun 16, 2019 8:33 pm
Here with scattering asymmetry down from 0.9 to 0.5 the light beam is stronger, but with classic engine the beam is blown out inside prism, don't know if the spectral rainbow will come out, perhaps it will. 55 passes.
55 Passes, multi scattering on.jpg
Prism BiDirVM.7z
Hi @Fox,
thanks, pretty helpful. Rendered your .blend with the new 2.2 beta version last night for an inital test but unfortunatly the prism is just a white splotch without the fine detail seen in your render. Do you have any idea what I'm missing here?
Cheers,
Christian
Re: Cold Fusion
Posted: Wed Jun 19, 2019 9:50 am
by epilectrolytics
@p2or:
That is probably the difference between BiDir and BiDirVM, which is not available from Blender.
You need to export the scene as a binary (Blender LuxCore render settings) which will produce some folders with a render.cfg file inside.
This you need to open with LuxCoreUI which is already installed in your BlendLux add-on.
When the LuxCore standalone is rendering, you can choose BiDirVM from the render engine tab which will restart the render with the new engine.
Re: Cold Fusion
Posted: Wed Jun 19, 2019 10:38 am
by p2or
epilectrolytics wrote: ↑Wed Jun 19, 2019 9:50 am
@p2or:
That is probably the difference between BiDir and BiDirVM, which is not available from Blender.
Oops. Of course, thanks @epilectrolytics
Re: Cold Fusion
Posted: Wed Jun 19, 2019 1:08 pm
by Fox
For classic engine it seems like smaller scattering value works better, or asymmetry value more forward scattering like 0.8 or 0.9
Then it's less white.
I got good speed up by moving camera closer to subject with wide angle lens 15mm from 70mm and making the scattering room / cube smaller.
Also i have the prism step size from 2mm to 1 cm and limited the max steps to 16. I get about 400k samples/sec, but still very noisy with BiDirVM at 2400 sp
Re: Cold Fusion
Posted: Thu Jun 20, 2019 9:50 am
by p2or
Fox wrote: ↑Wed Jun 19, 2019 1:08 pm
For classic engine it seems like smaller scattering value works better, or asymmetry value more forward scattering like 0.8 or 0.9
Then it's less white.
I got good speed up by moving camera closer to subject with wide angle lens 15mm from 70mm and making the scattering room / cube smaller.
Also i have the prism step size from 2mm to 1 cm and limited the max steps to 16. I get about 400k samples/sec, but still very noisy with BiDirVM at 2400 sp
Thanks @Fox!
I'll give it a shot and post my results using the
classic engine. Just noticed that you are using random sampling - any reasons for that? Speed? Besides the fact that your rendering already looking great, isn't metropolis sampling recommended for this kind of scene?
Cheers,
Christian
Re: Cold Fusion
Posted: Thu Jun 20, 2019 12:31 pm
by Fox
Perhaps the rough glass and clear glass mix material was causing noise with BiDirVM.
With simple clear glass
all the noise is gone, well 99,999%.
Re: Cold Fusion
Posted: Thu Jun 20, 2019 2:20 pm
by epilectrolytics
p2or wrote: ↑Thu Jun 20, 2019 9:50 am
isn't metropolis sampling recommended for this kind of scene?
There is a serious energy conservation bug in the BiDirVM+metropolis combo: the caustics get ever brighter with render time.
Sobol and random work (kind of) ok, but slower.
Re: Cold Fusion
Posted: Thu Jun 20, 2019 4:24 pm
by p2or
epilectrolytics wrote: ↑Thu Jun 20, 2019 2:20 pm
There is a serious energy conservation bug in the BiDirVM+metropolis combo: the caustics get ever brighter with render time.
Sobol and random work (kind of) ok, but slower.
Aha, good to know. Thanks @epilectrolytics