Am 30.12.2012 07:14, schrieb Kenneth:

> It raises a question in my mind, though: Given a particular math equation on an
> unchanging set of values, are the floating point errors always the same? As in,
> frame-by-frame animation?

Depends.

The floating point errors are deterministic, so with one and the same 
POV-Ray build they are always the same from frame to frame.

However, they may differ between builds, depending on floating point 
compiler settings; for instance, the old x87 instruction set allows for 
either 64-bit or 80-bit interim results, while SSE2 always uses 64-bit; 
and depending on optimization settings, very small numbers may or may 
not be truncated to zero. So if you split up an animation on a 
heterogenous render farm, results may differ between frames. (This is 
true for both SDL math and the actual render math.)

Post a reply to this message