Compiling ifort with AMD machines

[mergulhador]

Hi all,

We have two AMD boxes with Fedora Core and ifort as the main
compiler. One box is old 32-bit and the other is a shiny
64-bit opteron.

If you code does not address much memory (it only
needs cpu power), is there any real advantage to compile it using
one particular box ? In other words, should we expect *much*
performance differences form these two scenarios :

a. compile the code on the 32 bit machine and execute it on the
64 bit.

b. compile and run the code on the 64 bit box.

We appreciate any comments or directions that may lead
to these estimatives.

Thank you.

[TimP]

Compiling with either the 32- or 64-bit compiler running on the 64-bit machine should be faster. Run-time performance of the 32-bit build will not be affected by whether you built under 32- or 64-bit linux. Many applications, within the very broad category you describe, will run 10% faster (or less) when built with the 64-bit compiler, assuming you choose equivalent options (-xW) in the 32- and 64-bit builds. If your 32-bit box is old enough to require -xK, you may see a big advantage for -xW on a machine which supports SSE2.