Difference between LC pre-filter and render-time filter?

[Tommy_TGCG]

Hello!

I'm trying to set up an animation with moving objects with LC as the primary GI engine. I know this is not a way to get accurate GI, but this time I need speed .

As suggested by some forum experts, I'm using a high LC pre-filter values (around 100) to get rid of the flickering. And here's the question:

Why use pre-filter and how is it different from the usual "nearest" render-time LC filter? Both show similar results (at a "100" setting), and the render-time filtering is faster, because it's multithreaded.

Is there something that makes the pre-filter a better choice? The manual is not very precise on this subject. It might make a difference when we use the precalculated flythrough LC (I think pre-filtering only takes place once) but as for frame-by-frame GI i do not see the point?

[^Lele^]

Don't quote me on this, but on an X2-4200 (slow cpu) pre-filtering leads to better speed at rendertime, as the filtering hasn't got to be evaluated per shaded pixel.
Maybe with many cores the filtering @ rendertime leads to faster processing.
As for results, the prefiltering is always based on the nearest N samples, so it should lead to identical results when the rendertime filtering is set to nearest.

[Tommy_TGCG]

This may be so - I'm using two quadcores in a DR setup, hence the speed difference.

But besides that, thanks for clearing the question about similar end results of the two options.

[vlado]

The idea of the prefiltering is that if you used, say 100 samples at render-time, and used an irradiance map/brute force GI for the primary bounces, then looking up all the 100 samples for each primary GI ray is going to be very slow. So you can prefilter with 100 samples instead, and then look up only 5 or 10 samples during the rendering, and still get the same result with faster render times.

Best regards,
Vlado

[^Lele^]

Err, yes, I forgot to mention that my tests had Direct Computation GI as primary method, and I was checking for a super accurate result (on a cornell box of sorts).
The speed differences between the prefiltered and the other method were quite sensible.