Re: Announcing Intel Fortran Compilers 9.0

hchu@xxxxxxxxxxxxxx wrote:
>>I've been using the Intel EM64T (Linux) compiler without problems on
>>Athlon64 and Opteron processors since November.
>
>
> Do you really see any speed improvements? When we first purchased the
> Opteron cluser, we tested the combination of gcc and Absoft fortran.
> The results were very disappointing. Our program ran twice as slow
> as the same program compiled with 32 bit compilers running on Windows.
> We even considered purchasing Pathscale compilers, but the process
> is pain in the neck so we stayed with Intel 32 bit compilers v.7(we
> didn't renew the support contract in time).

I certainly saw improvements; but one has to be specific about what is
being compared with what.

In my case, the 'old' machine was a 32-bit Athlon 2600+, and the new
machine was a 64-bit Athlon64 3000+. Although I think AMD's speed
ratings are naff, they provide a good indication of the relative
performance of their CPUs, and indeed the speed increase I saw was
approximately 3000/2600.

To be honest, I didn't look into the issue a great deal, but I didn't
find a huge amount of difference between code generated using the 32-bit
versus 64-bit compilers. I was expecting to see more of a differential,
on the grounds that the 64-bit compiler would generate code that could
take advantage of the new general purpose registers. Maybe my codes are
more limited by memory bandwidth?

Having said that, I find very significant differences between ifort, g77
and Lahey's lf95 -- ifort generates code that is consistently twice as fast.

I'm surprised that you got disappointing results on your Opteron
cluster. You say your program ran twice as slow, although you don't say
relative to what. However, can I ask whether your program is well suited
to something like an Opteron cluster? Is it parallel in any sense? If
not, you won't see any improvements over a single-CPU machine -- apart
from the fact that (with appropriate scheduling software, such as Sun
Grid Engine) you can run multiple, independent instances of the same
program -- maybe as part of a search through parameter space.

cheers,

Rich
.

Relevant Pages

  • Re: Fortran compiler for Opteron with OpenMP
    ... > Intel does not claim that its compilers support Opteron. ... We have used 64-bit versions of Absoft, Portland Group Inc, and NAG compilers ...
    (comp.lang.fortran)
  • Re: Announcing Intel Fortran Compilers 9.0
    ... > I've been using the Intel EM64T compiler without problems on ... > Athlon64 and Opteron processors since November. ... as the same program compiled with 32 bit compilers running on Windows. ... Previous versions of IVF have been disappointing. ...
    (comp.lang.fortran)
  • Re: Cray-pointer-based array initialisation
    ... PGI compilers 5.2 and 6.2 with the ACML and the december ... version of the Sun Studio 11 compilers + sunperf. ... Opteron processors. ...
    (comp.lang.fortran)
  • Re: Announcing Intel Fortran Compilers 9.0
    ... >>support for AMD Opteron is now claimed ... > in your C++ compilers? ... to 64-bit Athlon CPU; which also required a new motherboard; and a new ...
    (comp.lang.fortran)