From: "Robert Duke"
Subject: Re: AMBER: INtel Fortran 90 compiler
Date: Wed, 24 Mar 2004 15:36:35 -0500

Carsten -
I presume you are talking about on Linux IA32 systems. The short answer, in
my opinion, is NO. Currently, at least for pmemd, using the intel fortran
compiler is a bit of a mess because the latest version is broken in regard
to its ability to compile pmemd, though the rest of amber 8 may well
compile. This is NOT a pmemd problem; the compiler croaks attempting to
compile source that works fine for about half a dozen other fortran 90
compilers and all previous versions of the intel fortran compiler through
7.1.038 (they just broke 7.1.040 too; its a simple fix, but just use 7.1.038
instead). I have been told by intel that a fix has been done in
development, but they have not released it yet. Why use the intel fortran
compiler, given this grief?
1) It is clearly faster than all the competition (that is, it produces the
fastest code, by up to a factor of 2).
2) It is free, in an unsupported version (more like limited support, really)
to noncommercial interests (eg., academics). I believe there is a GPL
version of f90 in the works, but last time I checked it really was not there
yet, and I would be very surprised if it can beat or match the performance
of the intel fortran compiler. Who was the idiot that led the amber charge
into the world of f90, leaving behind good old g77? That would be me, for a
whole bunch of good reasons ;-) Unfortunately, we just really got burned by
this intel compiler breakage mess, right around the time of a major release,
and hopefully it will all be better soon. It is really a shame; I have used
versions 5, 6, 7, and 7.1 without any real problems, and now this...

SO, in the meantime, what to do?

1) Sign up for a paid-for or noncommercial intel fortran compiler license at
www.intel.com.
2) You will download the version 8 compiler first, whether you want it or
not (this is what you are given the choice of doing). You will also end up
setting up an account for premier support and be given a link to
https://premier.intel.com. If you go to this link and log in, you can then
download older versions of the compiler. For now, if you are on redhat 7.3,
8 or 9, or probably 3.1 RHEL, you should probably download 7.1.038, which
supports glibc 2.2.5, 2.2.93 and 2.3.2 (though the release note does not
claim this). It is a snap to install. If you have an older redhat, you may
have to check the glibc version, and then use an earlier release of 7.1 that
supports it. I happen to use redhat 7.2 a lot, which has glibc 2.2.4, and I
use an older version of ifc (build 20030307Z, built March 7, 2003) with that
system - this may be the original ifc 7.1, but I don't remember. It may not
be necessary to use an earlier version, but I have not checked it out yet.
3) If you are using later versions of Redhat linux, and especially if you
are going to use mpi, be sure you have applied all the updates available.
The linux guys have made a mess of the threads stuff in the latest releases,
with all sorts of thread stack problems manifesting themselves, especially
if you are not completely updated. There are comments about this in the
intel fortran release notes; I don't know if following the various
suggestions there are necessary, as I have yet to get a system with glibc
2.3.2 running under mpi.
4) Beware of RHEL 3.0; update it to RHEL 3.1 at your earliest convenience.
I have spent (wasted?) a lot of time trying to get things running on a dual
processor opteron with RHEL 3/3.1. Ultimately, I decided that I was having
problems with the chips overheating, rather than some obscure threads
problem, once I had the system completely updated and using ifc 7.1.038.
Prior to the recent release of 7.1.038, these systems with RHEL 3/3.1 were
completely useless as far as I am concerned (pmemd guy). I was very excited
about the dual opterons; they are very fast, but the one I got was not
stable with MD software running. I finally decided that the most probably
problem was heating when I also had problems with sander 7 compiled with g77
(md calcs really heat up the chips due to all the fp calcs). I never got
anywhere near running mpi on this beast, and finally sent it back. I would
appreciate it if everyone shares their positive and negative experiences
with these chips. Athlons have always worked great for me.

Sorry for the long mail, and the mess. Let me know if I can do anything
else to help. I will let everyone know when ifort 8 works with RHEL 3.1 on
dual pentiums, +/- mpi.

Regards - Bob Duke

----- Original Message -----
From: "Carsten Detering"
To: <amber@scripps.edu>
Sent: Wednesday, March 24, 2004 2:03 PM
Subject: AMBER: INtel Fortran 90 compiler


> Hi Ambers,
>
> Is there an easier way to obtain and use a fortran 90 compiler that the
> Intel one?
> I am installing Amber8.
> Thanks in advance,
>
> Carsten