Same scene, different render times

[PastaJackal]

Hi all,

I've been scratching my head over this for the past few hours. Same scene, but the camera which is sitting outside the doorway takes ~8 times longer to render. This is for an animation so really need some suggestions as to why this might be happening.

Tried so far:
-made sure the ivy is not using an alpha (all geo) and has a blur value of 1.0
-glass is using thin refractions without any volumetric colouring

I've tried the sampling focus pass to get more information but it didn't really give clear results.

As for lighting, I have sun and sky for the environment, large filler plane lights along the interior, lots of spot lights without IES, a few sphere lights for pendants, and I'm using CoronaLightMtl with emit light turned off for the inside of the spotlights etc. LightMtl behind the shop logos has emit turned on but that is the only one used.

Any ideas why this is such a drastic difference? Thank you!

[TomG]

My first thought on testing would be to look at what is different, and that's the ivy - I'd try hiding that and rendering the outside view and see if it makes the render time more in line with the other scene (which doesn't have the ivy visible). The fact that the ivy is all geometry does sound like it could slow things down.

Shame the stats aren't visible on the internal render to compare what you are getting on those (if you can post, could be interesting!)

If it's not the plants, then maybe the glossy pink floor (also not visible in the first image). You could also try a material override and see if that makes the huge difference in render time, in case it isn't the heavy ivy geometry.

[Njen]

Generally speaking, most renderers take longer to render darker images. While I haven't tried this trick with Corona yet, but in the past, for dark scenes, I've boosted the exposure of all of the lights to get a faster and cleaner render, then in post, simply exposed down the output.

It mainly has to do with how renderers distribute samples, which is weighted based on a curve to attempt to be more efficient. Again, I'm not saying this is the exact case with your render in particular or Corona in general, but it's something to maybe test for yourself.

[maru]

First of all, what exactly do you mean that "one takes x times longer than the other"?
Does it take x times longer to reach the same noise limit?
Does it take x times longer to reach the same pass limit?
Does it take x times longer to reach similar visual quality?
...or maybe something else?

[PastaJackal]

First of all, what exactly do you mean that "one takes x times longer than the other"?
Does it take x times longer to reach the same noise limit?
Does it take x times longer to reach the same pass limit?
Does it take x times longer to reach similar visual quality?
...or maybe something else?

Hi Maru,

Yes, it took about eight times longer to reach 6% noise level than the other image with same settings. Have been on this again for several hours today - had some success with boosting the lights in the scene but can't get this entry door image below a 50minute render. For reference, the other image takes 20minutes with the same render settings.

I've since optimised the ivy to remove all translucency and enable clipping in the opacity channel. It doesn't bloat render times much on its own, it seems to be a combination of the whole entry facade (glass, neon lights - which are lightmtls with no emission etc)

[Bzuco]

Just move small region render during rendering to determine which scene part has lowest samples/s.

[PastaJackal]

Thanks, not a bad idea! Haven't considered using that feature.

Is there any future plans to implement a render pass with samples to better diagnose problematic areas?

[Njen]

Thanks, not a bad idea! Haven't considered using that feature.

Is there any future plans to implement a render pass with samples to better diagnose problematic areas?

It already exists: "SamplingFocus"

[Bzuco]

Sampling focus just tells you where is adaptivity concentrating more or less, it does not tell you how hard is rendering some material for corona.

[Njen]

The two are related, more samples means more effort.

[Bzuco]

Adaptivity is sadly still unfinished feature from corona 1.4, so if you want to know what really caused 8x bigger render times you need to know which material is fastes to render and which slowest to render. Sampling focus render pass just tell you, how smart or how dumb adaptivity is.

PastaJackal: try disable adaptivity, maybe you will hit target noise level with less render time. Also If you decide to optimize materials, start with material which cover biggest area in your scene ( pink floor ). Please post here sampling focus render element, it is possible, that adaptivity is spending most time on material which is hardes to render ( lowest samples/s ) thats why you see so big render time difference ...maybe I am wrong, we will see.

[Bzuco]

From your render stats I see, that your scene has 52M unique polygons. That starts to be a lot. If you can optimize some objects which have lots of polygons, do it (reduce polygons on plants stems, flowers, maybe leaves...also rounded objects ). It helps you overall increase samples/s and hit that 50min. render time.

[Njen]

Adaptivity is sadly still unfinished feature from corona 1.4, so if you want to know what really caused 8x bigger render times you need to know which material is fastes to render and which slowest to render.

It's not just materials, but lighting as well. The attached images are an example, the second two images have a plane in between the light and camera that is bouncing light down on the objects:
img1 - light intensity: 1000, bounce plane: hidden, render time: 1:21, passes: 385
img2 - light intensity: 1, bounce plane: hidden, render time: 1:28, passes: 425
img3 - light intensity: 1000, bounce plane: visible, render time: 2:07, passes: 615
img4 - light intensity: 1, bounce plane: visible, render time: 2:47, passes:825

As you can see clearly, brighter and direct lighting results in faster renders, darker and indirect lighting results in slower renders.

Please post here sampling focus render element, it is possible, that adaptivity is spending most time on material which is hardes to render ( lowest samples/s ) thats why you see so big render time difference ...maybe I am wrong, we will see.

Err, that's what I was suggesting in the first place...

[Vlad_the_rant]

Has this issue been resolved? Did you find out what was slowing down the render? If you have any more trouble, please let us know.

[PastaJackal]

Thanks everyone!

Unfortunately due to time constraints on this project, I could only try several of these solutions before moving on. It was still tricky to discern what was causing the blow out in render time. I ended up having to set the noise level to around 8% and relying on denoising to get the animated frames out. I would really love a more comprehensive tool within Corona to analyse what materials might be troublesome, similar to what vray has.