Page 1 of 1

Better render settings for this object/material

Posted: Sat Jul 06, 2019 11:44 pm
by Headroom
This is a complex object and even when rendered with "normal" materials takes a while to "clean up". However, nothing beats the visual complexity of glass materials with couchy B. The object is lit by a 4k HDRI form HDRI haven called the Shanghai Bundle

As can be seen in the image below there still is a lot of color noise present using Bidir + MLT.
Screen Shot 2019-07-06 at 7.24.05 AM.jpg
Screen Shot 2019-07-06 at 7.24.29 AM.png
Screen Shot 2019-07-06 at 7.39.58 AM.png
Is Bidir MLT really the best render mode for this ?

Are there render settings that can be improved upon for such a scene ?

What SPP might be enough for a clean render ?
I submitted this to the Ranch Render farm and even 1000 SPP is not that much better.

Re: Better render settings for this object/material

Posted: Sun Jul 07, 2019 9:02 pm
by Oscar J
Would have suggested blasting it in GPU mode, but you probably need a lot of path depth.

I'm not sure you need MLT for a scene like this, as there's no particularly hard caustics to solve due to the fairly flat lighting. Tried BiDir without MLT?

Re: Better render settings for this object/material

Posted: Sun Jul 07, 2019 11:11 pm
by Headroom
No, actually I was convinced that I needed MLT. But after 16 minutes of rendering in BDIR it looks promising. I've also given it 17 minutes in Path/GPU and that also does not look too bad.

It just takes a little patience to render this on a 2017 MacBook Pro ;-)

Re: Better render settings for this object/material

Posted: Tue Jul 09, 2019 7:42 am
by FakeShamus
the thing about straight BiDir Pathtracing is it tends to keep clearing up progressively. MLT will show you the pretty caustics sooner, but the overall noise level does not decrease in such a linear way. I used to swear by MLT for everything, but Indigo's normal PT has gotten super fast.

Re: Better render settings for this object/material

Posted: Wed Jul 10, 2019 6:20 am
by schwungsau
you using GPU on a macbook ? wow... i always afraid it i push the GPU on laptop to hard to will burnout after some weeks/month... the macbook GPU's are generally some speed as the CPU... depending on your CPU, PR may be faster on CPU.

Re: Better render settings for this object/material

Posted: Wed Jul 10, 2019 8:40 am
by Headroom

I have to admit that I am uneasy about using GPU on a mobile platform ever since after OscarJ mentioned having serious problems with failed solder joints on the GPU on his earlier generation MacBook.

The model was just an exercise to see how far I can take things with the array modifier in Blender and then I decided to see how Indigo would render this.


Some of this is for me to figure out what render machine I want to invest in. An external GPU would be cost effective option if pure path tracing can do what I need. GPU path tracing creates magnitudes higher samples per second, but is that enough for complex scenes ?

I dug out a super simple Indigo test scene I created many years back that I believe would be very difficult for path tracing.
The image below was rendered using bidirectional path tracing without MLT to 10000 SPP (on the Ranch Render farm). It is clearer but there still is a good bit of color noise visible.

Re: Better render settings for this object/material

Posted: Wed Jul 10, 2019 9:09 am
by FakeShamus
10000 spp is a lot...
yeah, that's just a crazy complex image to resolve with so many reflections and refractions. I wonder if something like photon mapping would eventually help this out - I believe that was once in the works as an additional rendering option, it but I don't know what happened to it.

Re: Better render settings for this object/material

Posted: Thu Jul 11, 2019 2:34 am
by Headroom
So here is a bt more food for thought. Below are 4 images of that old test scene. Its basically an Aluminum 'pill' with water in it and a glass window. the Glass is simple glass is a specular transparent material with IOPR of 1.5 and no Couchy B. The ware material is this from the online material DB
Th light source is the catured simulation sun.

All of these images were rendered for 30 minutes on this machine;
Screen Shot 2019-07-07 at 9.21.10 AM.png

The path tracing with MLT render appears to be the best solution for this scene and it clears faster than BIDIR with MLT. I suspect that BIDIR MLT is better when there are multiple small light sources involved (and caustic es, of course).
I need to verify that with another scene.

The BIDIR CPU render is almost completely bare of any caustics, but in the Path tracing with GPU render caustics slowly appear. It would be interesting to see how fast that clears on a real GPU.