Having investigated a little further:

The artifacts relate to the 32x32 pixel 'blocks' into which the image is carved
for separate rendering by different threads (I guess there's a proper term for
those blocks?). It appears there's a random chance a block may erroneously come
up with a slightly bright result, consistent throughout that rendering block.

The problem happens under 64 and 32-bit windows versions, and Linux.

The problem persists even when using +wt1 to force it to run just one thread. In
this case it looks to me as though the distribution of brighter blocks may be
reproduced exactly the same on subsequent renders, though with more threads it
varies from render to render.

In my more complex scene, the problem persists even when I remove the pigment
from the differencing box (which seems to make it go away in the simple version
above).

Cheers,

Andrew.

Post a reply to this message