Compiler speedup with proper compiler flags

[postaquestion]

I have bought an 8-core (two quad)Harpertown machine from PSSC Lab. I used Intel compiler to compile my MPI-code which includes a lot of linear algebra and FFT's. I use MPICH to run it on 8-nodes. It runs 2 times slower than my colleague's similar AMD machine even though my machine processor frequency is 3.0Ghz while AMD has 1.6Ghz. The vendor of my machine believes that if I will use proper compiler flags I can substantially improve the performence.

Could you please suggest me what kind of flags can improve it for your quad Harpertown machine?

[Steven_L_Intel1]

Could it be that your application is memory bandwidth limited?

Try /O3 /QxT to begin with.

[TimP]

MPICH covers a wide range of possibilities. I assume you mean the identical binary one is used on both machines. Even so, a non-optimum MPI could make the difference. I have not seen an MPICH which worked reliably on quad core (unless you count Intel MPI or maybe Microsoft, which are based on MPICH). The 32-bit MPICH was never tested on dual or quad core, since those weren't available when it was made, to my knowledge.
For linear algebra and FFT, using an optimized library on one machine and compiling from source on the other would easily make the difference. Presumably, you are performing operations covered by Intel MKL library.
I assume you got Harpertown with at least 1333 Mt/s FSB, but it is certainly possible to construct a memory limited benchmark where the 1600 Mt/s FSB would be required to match the most recent AMD machines.