The amber PC cluster page has not been updated since last
September/October but the information should be still more or less
valid. As Jarry noted, the problem almost always resides in a
compiler/mpich combination. Serial versions of amber7 work with most
compilers. Amber7 was not well tested with intel ifc compiler (serial
should work fine and fast, but you may encounter problems with parallel
runs: from my experience, ifc compiles amber7 with mpich but crashes
when running the executable). Therefore, I recommend to stick to
g77+mpich combination for parallel runs. Even then, I recall having
problems on RH8/9. They were related to compilation problems of mpich
with new g77/gcc-3.x series, which was introduced for the first time in
RH8. However, I need to verify this because I don't remember which
version of mpich I was dealing with at the time. In summary,
RH7.3+g77+mpich is a safe combination for amber7. For the future, I'd be
focusing on linux distributions which are supported by intel compilers
(their latest release is version 8 and supported glibc libraries and
distributions are listed on intel wesite as well).
Cheers,
-Viktor
>this info is getting a little out of date, but we haven't
>modified it since lately we've mostly been using
>the upcoming amber version 8 on our machines. we'll
>be updating it soon to reflect our experience with the
>new code.
>
> > Has anyone else experienced these problems and is there a way
> > around them expect using only the serial version? Thank you
> > very much.
>
>We are recently seeing the exact same thing on a SuSE 8.2 system
>using gcc 3.3 and mpich 1.2.5. I was going to try a different
>compiler and see if that helps (Portland or Intel).
>
>Your reports of Redhat 8/9 working ok surprise me a little as the
>Amber web site indicates observed problems there, also.
>
>http://amber.scripps.edu/cluster_info/index.html
>
>Hopefully, it's a matter of getting the right combination of
>compiler and mpich. I'll let you know if we get something
>that works.
>
>Jerry
>
>-----------------------------------------------------------------------
>The AMBER Mail Reflector
>To post, send mail to amber.scripps.edu
>To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
>
>
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Sat Jan 31 2004 - 01:53:00 PST
