On Apr 16, 2009, at 5:57 PM, Kirk, Benjamin (JSC-EG311) wrote: > So have you learned your lesson and gone to gcc-4.3.3 on OSX?
No way... GCC is horrifically slow on OSX. It takes _3_ times longer to solve a problem compiled with GCC as opposed to Intel.... and that's not doing anything fancy with the Intel compiler. And that doesn't depend on which version of GCC at all. Currently, I'm using GCC 4.4.0 (pre-release) with auto-vectorization.... it does the best... but is still 2-3x slower than Intel. Really. I know it's hard to believe... I didn't believe it for a long time either.... I've talked to Apple about this... and they don't understand either. We first started investigating because Linux on the same machine was twice as fast! For a long time we thought OSX was just a dog... but it turns out to be some interaction between OSX and GCC.... Doesn't matter much though. Buying a few floating licenses of the Intel compiler suite is chump change next to the Dual-quadcore 2.93GHz Nehalem Mac Pro I just bought (yes, with hyperthreading it shows up as having _16_ processors.... and actually acts like it has about 12-14 in it). ;-) Derek ------------------------------------------------------------------------------ Stay on top of everything new and different, both inside and around Java (TM) technology - register by April 22, and save $200 on the JavaOne (SM) conference, June 2-5, 2009, San Francisco. 300 plus technical and hands-on sessions. Register today. Use priority code J9JMT32. http://p.sf.net/sfu/p _______________________________________________ Libmesh-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/libmesh-users
