GCC vs Intel CC

[elam]

I'm sure I'm making some false assumptions here, but shouldn't a 64 bit build always render faster than a 32 bit? For my build of 0.5,X86_64 I simply used -march=athlon64. I have AMD 64x2 4000+ with 4 gigs of ram.

And why is does the cvs gui look different than the luxrender.net builds?


X86_64 from luxrender.net:

[2 threads] 00:00:05 18458 samples/sec 18458 samples/totsec 0.156073 samples/pix
[2 threads] 00:00:10 18128 samples/sec 18300 samples/totsec 0.296309 samples/pix
[2 threads] 00:00:15 17283 samples/sec 17971 samples/totsec 0.430006 samples/pix
[2 threads] 00:00:20 17725 samples/sec 17911 samples/totsec 0.567124 samples/pix
[2 threads] 00:00:25 17947 samples/sec 17918 samples/totsec 0.70596 samples/pix
[2 threads] 00:00:30 17927 samples/sec 17920 samples/totsec 0.844639 samples/pix

X86_64 my build:

[2 threads] 00:00:05 19256 samples/sec 19256 samples/totsec 0.169527 samples/pix
[2 threads] 00:00:10 19633 samples/sec 19432 samples/totsec 0.321401 samples/pix
[2 threads] 00:00:15 18274 samples/sec 19063 samples/totsec 0.462767 samples/pix
[2 threads] 00:00:20 19580 samples/sec 19188 samples/totsec 0.614231 samples/pix
[2 threads] 00:00:25 17669 samples/sec 18892 samples/totsec 0.750917 samples/pix
[2 threads] 00:00:30 18730 samples/sec 18866 samples/totsec 0.89581 samples/pix

Vista 32 bit from luxrender.net

[2 threads] 00:00:05 21062 samples/sec 21062 samples/totsec 0.175904 samples/pix
[2 threads] 00:00:10 20955 samples/sec 21010 samples/totsec 0.338008 samples/pix
[2 threads] 00:00:15 20598 samples/sec 20877 samples/totsec 0.497355 samples/pix
[2 threads] 00:00:20 21018 samples/sec 20911 samples/totsec 0.659951 samples/pix
[2 threads] 00:00:25 21207 samples/sec 20970 samples/totsec 0.824006 samples/pix
[2 threads] 00:00:30 20517 samples/sec 20895 samples/totsec 0.982725 samples/pix

[Dade]

I'm sure I'm making some false assumptions here, but shouldn't a 64 bit build always render faster than a 32 bit? For my build of 0.5,X86_64 I simply used -march=athlon64. I have AMD 64x2 4000+ with 4 gigs of ram.

A 64bit build should outperform a 32bit most of the times (mainly because of the additional registers available in 64bit mode, etc.). The problem is you are comparing apples with pears: you should compare a "-march=athlon64" 32bit build with a "-march=athlon64" 64bit build.

Your test shows Visual C++ compiled code is faster than GCC compiled code (and not that 32bit code is faster than 64bit).

And why is does the cvs gui look different than the luxrender.net builds?

The old GUI is still named luxrender on the CVS while the new GUI is named wxluxgui. wxluxgui has already replaced the old one in luxrender.net builds.

[Rom1]

Hi there,

I'm looking for some optimization of my luxrender.

Did you know if the optimization with intel compiler are still available ?

I use MSVC and I want to reduce simulation time. I can't use GPU for the moment, except if I can found a graphics card with 12GB GPURAM ^^

[SATtva]

I can't use GPU for the moment, except if I can found a graphics card with 12GB GPURAM ^^

Maybe Teslas?

[Rom1]

Maybe a little bit expensive ^^

and only 6GB

[Rom1]

hi there

I'm trying to compile with ICC ...

I use VS2008 and my lux compile properly

I get a trial version of ICC, with some effort I successfully compiled boost with it.

Now I want to compile luxrender ...

I changed "use intel c++"

But impossible to have luxrender working

I tryed manually editing the vsprops, because it can't finf some variables.

It work, it compile, I get a luxrender, but when I open a .dis it can't find any of the "plug-ins" ...

I tried to replace all boost-vc90-lib by boost-iw-lib with modified vsprops files ...

No way I get a build but it doesn't work, I have a unhadled exception ....


Please help !

[jeanphi]

Hi,

You need to include all symbols from the lux library at the final link stage, I don't remember the option for that with icc.

Jeanphi

[Rom1]

I'm sorry

I tried a lot of things but I can't have it working