I just ran a few comparisons with 3.7.0-unofficial for Mac.

The first test, a pretty simple model, seemed to run significantly faster on
3.8.0.  So I tried a quite complex model.  At quality level 9 but a 320x180
output to keep run time reasonable, I see the following:

3.7.0: 5:52 total, 2727 seconds trace time (8 threads), 741770 rays
3.8.0: 4:42 total, 1829 seconds trace time (8 threads), 786430 rays

So 3.8.0 is doing a bit more work (more rays, and the detailed work listing also
show more intersection tests etc.) but it does it substantially faster.

Naturally, I don't mind, but I thought it was curious, there's nothing I saw in
the changes list that suggests this is expected.   Could it just be a better
compiler?  3.7.0 mentions  Compiler: gcc 4.2.1 Compatible Apple LLVM 6.0
(clang-600.0.54) -- the 3.8.0 message window omits that information.

There is no obvious difference in the generated output images.

Post a reply to this message