Am 13.06.2017 um 09:07 schrieb dick balaska:
> Am 2017-05-18 12:26, also sprach clipka:
>>
>> Performance seems to be just as good as the native Windows binaries --
>> as a matter of fact, on my machine the Unix variant of beta.7 (compiled
>> with g++ 5.4.0) comes out slightly ahead by about 1%.
> 
> I have one thing I've been wanting to write up...
> In the kitchen of my animation, the tile floor is superellipsoids with a
> box for grout between the tiles.
> When the camera looks at this floor, Windows renders this 10x faster
> than Linux/gcc [1].  (The grout must be present for the extreme
> variation.) At least my theory is it's windows vs. Linux.  The problem
> with my theory is that my Windows box is also my only hyperthreaded i7
> cpu. My linux boxen are all i5(s).  But I don't think it's the CPU, I
> think it's the compiler.  Anyway, now I have to write this up as a
> proper test. I thought it might be interesting to compare against the
> clang results, too.

This would have to be examined in more detail.

Do you still see this speed difference with current beta versions?


With "Windows Subsystem for Linux" now able to compile and run POV-Ray,
it should be easy to figure out whether this is a matter of the CPU or
the compiler: Just run POV-Ray for Windows and POV-Ray for Unix on the
same machine. If the factor 10 speed difference is absent, the compiler
is free of guilt; if it is present, it must be the compiler that makes
the difference, and we might want to figure out how to get better
performance out of g++ for whatever piece of code is affected by this.

Also, it might be interesting to compare the performance on different
CPUs, to see whether the Unix performance loss is a general problem or
limited to a particular class of machines. We're in contact with some
folks at AMD who might be able to test-drive a sample scene across a
wide range of different machines (both AMD and Intel).

Post a reply to this message