On Thu, Feb 17, 2005, Greg Recine wrote:
> We have a netBSD cluster and would like to run Amber8 on it.
I am *not* running any flavor of BSD, so take my comments with a big grain of
salt:
> (3) use gfortran to compile Amber. This would be great for both my
> netBSD and Linux boxes.
My experience so far with gfortran on Linux and Windows is that it is not yet
ready for Amber. (Last version I tried was from 17jan05; a more recent update
has been posted.) However, I have had good luck with g95 (www.g95.org) on
Windows and Linux. All of Amber 9 (development version) runs under g95.
There are some fixes that would need to be back-ported to Amber 8, but the
list is not all that large (especially if the goal is just to run sander). In
addition to source code, the G95 web site lists a FreeBSD binary package. So,
it seems likely that the compiler could run on netBSD as well.
As you might expect, however, the performance of G95-compiled code is below
that of Intel's ifort, at least on Linux. So, if your primary interest is in
high-performance, parallel executaion, this might now be such a good option.
(For sander, G95 code seems roughly 30% slower than ifort on Linux on a
Pentium IV single cpu; for pmemd, the "G95 penalty" is greater, 60% or more.)
Of course, no "tuning" of Amber for G95 has been done at all.)
I have *not* used G95 in parallel on any machine. But it seems likely that
mpich or lam compiled with the gnu compilers would work in combination with
G95.
And, as you point out, it might be more straightforward or productive
to use ifort in an emulation mode.
....dave case
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Fri Feb 18 2005 - 07:53:00 PST
