"Cousin Ricky" <ric### [at] yahoocom> wrote:
> What really shocked me was the amount of time it took to render.  At 42 hours,
> it's easily, and by far, the longest render I've ever done.  I'm at a loss as
> to why, for such a simple scene with no media, no refraction, and no
> dispersion.  It uses radiosity, focal blur, and an area light, but based on
> past experience, I didn't expect them to slow the render by 3000x! (The render
> time was only 50 seconds otherwise.)

Awesome shot!

But the combination of radiosity, focal blur and area lights can really kill.

My advice:

- Do radiosity calculations in a separate shot (using the save_file/load_file
mechanism) WITHOUT the area light (i.e. just a point light) (and of course
without focal blur)

- Check whether you're possibly overdoing the focal blur

- Make sure you use *adaptive* area lights


Say you have set focal blur parameters to render up to 100 rays per pixel. Now
add a non-adaptive 5x5 area light (or an "adaptive 1" area light, which always
shoots 5x5 rays minimum as well) - and voila: Every pixel sufficiently out of
focus will take about 2500-fold the time required for a shot without focal blur
and area lights.

(Note the "Smpls/Pxl: 108.71" - I guess these are *initial* rays shot per
pixel.)

The weird thing about this is that with 100 rays shot per pixel, you wouldn't
actually need an area_ligt at all - if you could just jitter the light source a
bit for every ray.

So you might actually achieve the same quality 30 times faster, by doing 100
shots with some suitably jittered non-blur camera and a jittered point light
source, and finally generate an average of all the 100 shots.

Post a reply to this message