Understanding render times - Sampling Focus

[romullus]

Yep,refraction in foliage material is sign of really bad setup. It's definitely worth to double check every material and turn refraction off.

[cecofuli]

Try to use CoronaRaySwitchMtl for the foliage and remove the GI from the leafs.
Maybe you can speed up a bit.
And, yes, Opacity vs Geometry you can speed up a lot!

Same amount of time:

[Bormax]

Working with exterior scenes I've faced the problem with long render times for the cases where I used HDRI map in Reflections override slot of Render setup window. It made render times dramatically longer for me. So don't use that, or you can use work around I found useful for me - use JPG version of the HDRI map I wanted to use. Yes, it kills some details in reflections, but I found some settings which can make reflections with JPG look almost like with HDRI map. If you want, you can find more information here - https://forum.corona-renderer.com/index.php/topic,13104.15.html

[Fluid_Jamie]

Thanks everyone, there's a few good suggestions here I need to try.

 I know someone mentioned earlier that there was a 'global opacity map filtering option', where would that be as I can't find it?

My render times are still higher than they should be, so going to keep trying!

[romullus]

It's in render setup>performance>dev/experimantal tab. If you can't find it, you have to unlock it in render setup>system>system settings. Just be warned to not mess up with anything you don't know about there ;]

P.S. filtering for opacity is disabled by default, so most likely your problems are somwhere else.

[Fluid_Jamie]

It's in render setup>performance>dev/experimantal tab. If you can't find it, you have to unlock it in render setup>system>system settings. Just be warned to not mess up with anything you don't know about there ;]

P.S. filtering for opacity is disabled by default, so most likely your problems are somwhere else.

Thank you, I thought I was going mad not being able to find it! It was disabled, but I had to check to be sure!

So yeah, my problems are elsewhere. I'll keep looking...

[maru]

Let's try some extreme optimizations.

1) Since there are some interior/gi parts visible, I would leave PT+UHD as solvers.
2) If there are no moving objects, just a fly-through, you could bake the UHD Cache once, and re-use it for each frame. This way you will save UHD calculation time for each frame. See: https://coronarenderer.freshdesk.com/support/solutions/articles/5000515648
3) Simplify foliage materials, unless close-ups or extreme realism is needed - remove bump maps, remove translucency, make sure reflections have low glossiness.
4) Set MSI in performance tab to 10 - this should reduce noise, and introduce slight bias, but it shouldn't be noticeable, especially in an exterior scene.
5) Set GIvsAA to 8, since more antialiasing may be needed for the foliage to look good. Otherwise flickering may appear on hard edges.
6) Set LSM to 1, and check if there are no issues like increased render times. If not, leave it at 1. If there are, revert to the default 2.
7) Set max ray depth to the minimum possible value. I would try with 5 and see if there are no issues like strange looking reflections or refractions.
8) Disable displacement if there is any.
9) Use denoising.
10) Use blur/sharpen. This will slightly blur the noise.
11) If the image does not have to be saved in 32-bit format for compositing, etc, change highlight clamping in the system tab (do not mistake with highlight compression!) from 0 to some low value like 2, and check if there are no issues. Set it to lowest possible value. This will clamp all brightest areas, including fireflies and noise in general.
12) Increase image filter width to 3. This will slightly blur the noise.
13) Use pass limit instead of noise limit. It will give you consistent results and probably render times per frame, even if the content of the frame changes.

Note that these are EXTREME optimizations, and generally this is something that is not advised to do with Corona. :)
If you will get some incorrect results, just revert to defaults, or previous safe setup.

[Fluid_Jamie]

Let's try some extreme optimizations.

1) Since there are some interior/gi parts visible, I would leave PT+UHD as solvers.
2) If there are no moving objects, just a fly-through, you could bake the UHD Cache once, and re-use it for each frame. This way you will save UHD calculation time for each frame. See: https://coronarenderer.freshdesk.com/support/solutions/articles/5000515648
3) Simplify foliage materials, unless close-ups or extreme realism is needed - remove bump maps, remove translucency, make sure reflections have low glossiness.
4) Set MSI in performance tab to 10 - this should reduce noise, and introduce slight bias, but it shouldn't be noticeable, especially in an exterior scene.
5) Set GIvsAA to 8, since more antialiasing may be needed for the foliage to look good. Otherwise flickering may appear on hard edges.
6) Set LSM to 1, and check if there are no issues like increased render times. If not, leave it at 1. If there are, revert to the default 2.
7) Set max ray depth to the minimum possible value. I would try with 5 and see if there are no issues like strange looking reflections or refractions.
8) Disable displacement if there is any.
9) Use denoising.
10) Use blur/sharpen. This will slightly blur the noise.
11) If the image does not have to be saved in 32-bit format for compositing, etc, change highlight clamping in the system tab (do not mistake with highlight compression!) from 0 to some low value like 2, and check if there are no issues. Set it to lowest possible value. This will clamp all brightest areas, including fireflies and noise in general.
12) Increase image filter width to 3. This will slightly blur the noise.
13) Use pass limit instead of noise limit. It will give you consistent results and probably render times per frame, even if the content of the frame changes.

Note that these are EXTREME optimizations, and generally this is something that is not advised to do with Corona. :)
If you will get some incorrect results, just revert to defaults, or previous safe setup.

That certainly is a very helpful list of suggestions, thank you! I am actually currently running some more tests on this project, and came back here to ask some more questions on tweaking the render settings. So your post was a convenient and nice surprise!

I'm glad you agree on the UHD situation, I settled on using UHD as it actually seemed to be performing better than PT + PT in my tests. I haven't seen any flickering in the short tests I have done, so I'm hoping it should be ok. This is only a flythrough, so I will precalculate locally before sending to a farm.

After all the suggestions on the thread, I have meticulously checked all the foliage materials, reducing reflection as well as removing bump and translucency. I honestly can't believe how many foliage materials used refraction, even forest pack materials incorporated 1.0 refraction! So that has helped quite a bit.

I'm currently playing with GIvsAA, my most recent test on 8 and LSM at default. While initial estimates suggest faster frames, it gets stuck and lower noise levels and reaches the cut off slower than when at 16. The sampling focus suggests it has having trouble with shadows? (See attached).

I'll try your other suggestions, but I'm certainly thankful for denoiser! I'm using a stable version of the Daily build (31/07/2017) which has helped render times too...

[Frood]

even forest pack materials incorporated 1.0 refraction!

Really? Which ones? Usually the refraction color is set to black in (library) materials. Not a good style, but no refraction in this case.


Good Luck

[Fluid_Jamie]

even forest pack materials incorporated 1.0 refraction!

Really? Which ones? Usually the refraction color is set to black in (library) materials. Not a good style, but no refraction in this case.


Good Luck

You're exactly right, the refraction is set to 100% black. I've been going through so many foliage materials in this scene I just saw the '1.0' refraction and got annoyed...

Thanks!

[Frood]

Let's try some extreme optimizations.

Another question Marcin: Is that vast amount of sampling in very dark areas where there is usually a lot of visible noise anyway (foliage with lots of fine geomentry) wanted? I'm aware of the fact that one may still increase exposure to some insane value and then it may become reasonable but this always looks a bit exaggerated to me since adaptivity was introduced.


Good Luck

[Ondra]

Let's try some extreme optimizations.

Another question Marcin: Is that vast amount of sampling in very dark areas where there is usually a lot of visible noise anyway (foliage with lots of fine geomentry) wanted? I'm aware of the fact that one may still increase exposure to some insane value and then it may become reasonable but this always looks a bit exaggerated to me since adaptivity was introduced.


Good Luck

we will fix this to make adaptivity faster in time for 1.7

[Frood]

Tnx Ondra, pretty curious about how it will be balanced.


Good Luck

[Noah45]

Appreciate the insane optimizing...under the gun, this is good for crazy deadlines.