I'm juggling my time between work arounds to meet some deadlines and trying
to get to the bottom of this.  I'll have to look into it deeper as time
permits, so I havent been able to explore all the helpful suggestions here
yet.

But one thing I'm seeing today is curious. I isolated one very small piece
of geomtery. Rendered it in native scaling, and it rendered in one second.
I scaled it by the .0000254 and it took 10 seconds to render (and I did
adjust the camera position so that the geometry covered the exact same
image portion.) I then scaled it to .254 and it rendered in 3 seconds.  The
slow down is porportional to the degree of scale, not just the scale itself.
Does that make ANY sense at all?

On a similar note, a light source I have was positioned at <-9500, 0, -9500>
and rendered fine.  But when moved to <-95000, 0 , -95000> * 3600, the
render hung, and I had to kill it.  Again, slowdown proportional to the
precision/magnitude of the values involved in the computation.

All this has been done on 32 bit architectures so far (linux and windows).
As of yesterday, I did a custom build on a 64-bit mega cluster. I've only
rendered my 'workarounds' on there so I havent learned anything yet, but
will shortly be playing with these issues in 64 bits to see if that sheds
any light.

I've just never seen this sort of thing before.

Post a reply to this message