Warp <war### [at] tagpovrayorg> wrote:
> Mark Weyer <nomail@nomail> wrote:
> > Povray isosurfaces are fastest when the actual gradient does not differ
> > too much from maxgradient.
>
>   You mean an isosurface which renders *correctly* is fastest when its
> gradient and the max_gradient are close to each other?

My statement was meant as a rule of thumb and is imprecise in several ways.
Furthermore it is only based on how I think it works.
My point is that I have found it worthwhile to get rid of 'nonlinearities'
in iso functions. Extra wrappers of sqrt or such might take more time per
sample, but make the algorithm require much fewer samples. Of course I am
only interested in correct renderings.


>   Naturally a lower max_gradient will always produce a faster isosurface,
> but it won't render correctly if it's too low, of course.

In this sense an incorrectly rendered iso renders fastest when it is omitted
(or maybe replaced by its contained_by -- it can even speed things up by
eclipsing complex objects) ;-)

I recall some SCC entries that use incorrect iso rendering on purpose.
But I am not aware of a 'serious' application of this practise...

  Mark Weyer

Post a reply to this message