Lower Subdivisions don't mean shorter render times!

[timmatron]

I was doing some tests today with reflection settings and render times and came across something interesting.

I thought & I think a lot of other people think that the lower your reflection subdivisions are, the shorter your render time will be....

I have a very simple scene set up. A plane, with a ball and a cylinder on top. The plane is what I am testing reflect settings on...

When I set subdivs. to 3 I get a render time of 14 seconds...

3 subdivs. = 14 seconds
10 subdivs. = 21 seconds
20 subdivs. = 10 seconds
50 subdivs. = 8 seconds
60 subdivs. = 9 seconds
100 subdivs = 14 seconds

So is there a sweet spot of 50?? Maybe that is why it is set as the default number?

Sorry, maybe some of you know this already, but I'm glad I did the test because I will end up saving myself a lot of render time in the future by not lowering subdivisions.

[RErender]

The subdivisions only come into effect when you are using Glossy Reflections (then you will definitely see a difference) the difference in time must of been due to other factors.

[timmatron]

Yes, I was using a glossiness value of .95 with Fresnel reflections checked.

[RErender]

Are you using interpolation? I am asking because if not on some scenes the difference between 3 and 50 could be days

[timmatron]

No interpolation.

See thats what I alway thought.. Higher subdivisions = higher render times. But thats not what this test proved to me.

Its a simple material with a bitmap in the diffuse slot and a bump map to match.

Reflection settings as stated before. Fresnel checked, glossiness at .95, no interpolation - all else default.

[RErender]

Perhaps a good secondary test would be two (or more) glossy objects reflecting eachother, the subs do have a large effect I'm not sure why it is not showing in your test scene.

[vlado]

Things also depend on your AA settings. Lower subdivs mean VRay will have to do more AA in noisy areas, which may explain the longer rendering times with lower subdivs.

Best regards,
Vlado

[timmatron]

Yea, I was starting to think about that. So the more noise that is produced from low Subdiv settings may mean longer times to do AA. Interesting.

I was trying my test some more with more objects of the same material. On one test the difference between 3 and 50 was only 1 second in favor of 3.

So I guess generally speaking, lower subdivisions do mean better times. But at the same time, I think there would be a sweet spot for every condition. yikes! guess its all about trial & error.

[timmatron]

Ok, this is the scene I've been testing. All objects have the same material applied.

I tried same rendering at different subdivision levels and found that for this particular scene, 10 subdivisions gave me the best time. Here are all of my results. The first number is the amount of subdivisions, the second number is the render time.

50 = 4:09
25 = 3:41
20 = 3:39
15 = 3:38
10 = 3:28
5 = 4:06

So, I'm sure each scene, material & reflections are completely different, but I think its still good to keep in mind that there is a point where lower subdivisions could actually cost you time.

[Geddart]

You tested it with adaptive AA right?

If yes, you have to do some more tests with simple two level and fixed rate. Many people don't understand that adaptive AA works good in Scenes with large uniform coloured surfaces but fails with scenes where a lot of per pixel effects are involved. (like your non interpolated glossy reflections/refractions) Adaptive AA tries to smooth out the per pixel mess (contrast) but fails cause the maximum number of samples is limited in the material. The result are loooong rendering times. (sometimes they're even longer as if you'd use a higher subdivs setting for your glossy effects in your Material)

This is the same thing when you have many small objects in your scene resulting in heavy contrast differences at a small scale. Use simple two level then. (or even fixed rate!)

[timmatron]

Thanks for the advice. I've usually used the adaptive AA just figuring it was typically the fastest. I'll do some more tests on 2 level and fixed rate.