> Did anyone try to reproduce the effect that 3.7 seems to be
> significantly slower (read: more than any reasonable overhead)
> on some single core systems (in my case: P4 2.6 WinXP)? This is
> of course not urgent for beta26, but if the final version has
> the same behavior, it will seriously deter some people from
> using the new version. I realize the real benefit of 3.7 is
> for multi-core systems, but it suppose it is also in your
> interest that all users work with the newest version.
> 
> For examples, rendering the sample advanced/abyss.pov with
> 640x480 AA 0.3 took about 3 minutes with 3.6 on my system,
> but 9 minutes with 3.7 in the SSE2 version using thread
> count 1.

In some features, 3.7 *is* slower than 3.6, for example due to global 
caches that were used before, but aren't thread-safe. It's not that big 
of a slowdown, though. Also, the tiled rendering used for multithreading 
causes extra antialiasing samples (on the tile edges). Is there a way to 
change the tile size yet?

Of course there were also optimizations to other features.

In any case, 3x slower doesn't look like just "overhead", that's a lot! 
If you see any important slowdown on a scene *without* media (see Warp's 
post), *that* would be a problem worth reporting.

Post a reply to this message