Vray Framebuffer improved

[jacksc02]

Aldaryn, what the direct rendering of caustics thing your talking about?

[rivoli]

I guess that's the one:

http://www.chaosgroup.com/forums/vbu...hlight=caustic

[Aldaryn]

Yup, thats the one.

Best regards,

A.

[^Lele^]

Vlado, I did a few tests (which i may publish for those interested), while in the process (stalled as of now) of writing a BSP optimiser script, to see how vray behaved changing the face/level coefficient, in terms of speed/memory usage.
I found out that after a certain level, decreasing that coefficient wouldn't improve rendertimes (which is of course well mentioned in the manual), while in theory it should have.
What i mean is that i presumed that over a certain point the renderer would swap to disk, making further optimisations in (average) number of triangles per voxel useless.
What i found out is that while it never swapped, that critical point held true.
For instance, for a given scene (many max 3d trees, not instanced), below an average of 5 faces/leaf, vray wouldn't gain any speed in the raytracing phase (be it LC, irmap, glossies and what not).
Also, lowering the coefficient further wouldn't result in big changes in the average number of faces/leaf, as it did before the critical point.
I would have expected, by sheer theory, that the "optimal" face/leaf level would have had to be 1 (ie, the raytracer only checks one triangle per ray, not having to waste time checking multiple triangles for intersection).

I just tried out the hidden option, and it seems to me that it's similar in behavior to lowering face/level coefficient, in that it splits space a tad more aggressively (actually, for the same face/level coefficient it seems to halve the number of average faces/leaf).
I haven't tested it extensively, yet, and the real question is this: what should i expect to see?
Does the new optimization lead not only to a finer degree of space partitioning, but will the tracing also benefit more from the lower number of faces to intersect?
In any case, if you find some time to spare, and should it not be an IP secret, could you elaborate more on the technicalities of this option?

Thanks in advance

[vlado]

First of all, the purpose of the bsp tree is not to determine which faces to intersect, but to throw away faces that don't need to be intersected. There is a balance between the computational cost of traversing a very fine tree, and the amount of calculations you save - a deeper tree does not necessarily mean any faster raytracing. Further on, there is a limit to how deep you can go - since the bsp tree is axis-aligned, it is nearly impossible to have only 1 face per leaf - in practice a single face will appear in multiple leafs. Beyound a certain point there is simply nothing to be gained from further subdivision.

By changing the ray caster optimization level, V-Ray tries to better estimate how deep to go, and how exactly to subdivide so as to get the best possible result. Also, in that mode, the face/level coefficient is ignored as V-Ray will use other heuristics to determine the optimal number of faces in a leaf.

Best regards,
Vlado

[^Lele^]

Nice, instructional and exhaustive.
Thanks a lot.

[priad]

What about a thing where you can adjust only the clamped colors, to unburn clamped area in some cases.
It would be very useful....

[Geddart]

Wow, after all those years!
Thanks a Lot for implementing the History Vlado.
I think with a few touchups it will be very handy.

[giraffe]

this is so great