Yes, absolutely right.  The memory requirements are only for 1000 spheres
and 4 CSG union objects.  In my ray tracer, any CSG object or triangle mesh
is automatically assigned its own individual subdivision scheme (any one of
uniform grid, bounding volume hierarchy or kd-tree ... although I've
never found any reason yet to use anything other than uniform grid ...
it seems to outperform kd-tree by at least 20% in almost all cases, with the
unfortunate side-effect of needing a little more memory), which makes
rendering a "huge" CSG about as quick as a similar-size triangle mesh.

You're right that this approach isn't particularly flexible but I've since
found it useful for simulating rough textures, such as that on the surface
of
a brick, by clumping together billions of virtual boxes (per above method)
with each group having a slightly different texture and size/rotation.

I liked the ethereal/huge-scale quality of some of the images resulting
from the experiment hence me posting it...

Ian.

Warp <war### [at] tagpovrayorg> wrote:
> IanT wrote:
> > I came up with a scene that effectively contains
> > a trillion (1,000,000,000,000) spheres.
>
>    Unless you are using a 64-bit computer with dozens of gigabytes
> of memory, I assume that you can't actually create that many
> individual and independent spheres (ie. so that the location and
> size of each sphere could be specified separately and independently
> of any of the other spheres).
>
>    If you just made a group of some millions of spheres and then
> referenced it (without copying) several times, then it's possible,
> of course, but you will only get copies of the group and not
> individual independent spheres.
>
>    Something similar to this can already be done in POV-Ray if you
> use triangles instead of spheres.

Post a reply to this message