POV-Ray : Newsgroups : povray.unix : Performance of GNU vs. Intel C++ compiler on AMD64 : Re: Performance of GNU vs. Intel C++ compiler on AMD64 Server Time
17 May 2024 10:16:27 EDT (-0400)
  Re: Performance of GNU vs. Intel C++ compiler on AMD64  
From: clipka
Date: 31 Oct 2009 15:03:02
Message: <4aec89e6$1@news.povray.org>
clipka schrieb:

> I'm just testing with g++ 4.3.2 now, and (as I would have expected) the 
> GNU suite has obviously improved. While it is still a tad behind on 
> average, it quite obviously plays in the same league now as the 11.0 
> version of Intel's compiler.

Some interesting side notes on this:

- GNU g++ 4.3.2 compiled POV-Ray is still comparatively slow for simple 
scenes that don't take long to render anyway; this may adversely affect 
performance when rendering animations (I didn't test it though, and it 
may depend on whether the slowdown is during program startup, parsing, 
or whatever).

- A /severe/ difference in performance can be seen with scenes using 
crackle pattern; for instance, the sample scenes "crackle2.pov" and 
"crackle3.pov" take twice as long with the g++ 4.3.2 version than with 
the icpc 11.1 version.

- Photons paint a totally different picture: When compiled with Intel 
icpc (both versions 11.0 and 11.1), POV-Ray has been segfaulting on me 
sporadically (but quite frequently) ever since, with any scene that uses 
photons. (We're talking about Debian Linux, both 4.0t5 "Etch" and 5.0.3 
"Lenny", running on an AMD Phenom X4 9650.) When compiled with GNU g++ 
(both versions 4.1.1 and 4.3.2), I have never encountered any such 
crashes so far. I actually suspect some bogus optimization here.


Post a reply to this message

Copyright 2003-2023 Persistence of Vision Raytracer Pty. Ltd.