Re: [AMBER] Installation on MacOSX

From: Jason Swails <>
Date: Tue, 15 Dec 2009 06:43:47 -0500


The latest intel compilers do work for Snow Leopard, and, as far as I
know, create 64-bit executables by default. The GCC compilers in the
latest Xcode Tools for Snow Leopard also create 64-bit executables by
default (this is not necessarily true of your gfortran, depending on
which binary you installed, but if you got one built for Snow Leopard
it will).

David's instructions will, I believe, build 64-bit executables for
AmberTools 1.2 and Amber10. I may be mistaken, but if you do not see
-m32 or -march=i686, or something to that effect in the compiler flags
(CFLAGS, FFLAGS, etc.) in the config.h file that is created by
configure, it will create a 64-bit executable. This is only
potentially an issue for Personal Macs (anything except Xserve with an
intel processor), since they boot a 32-bit kernel by default, so uname
-m returns i386, which many configure scripts interpret as a 32-bit
processor. This can cause problems if the program you're trying to
install attempts to link to existing system libraries, since they're
likely 64-bit and will thus be incompatible. (Holding 6 and 4 down at
boot boots into the 64-bit kernel, and uname -m will return x86_64 for
those Macs capable of doing that). Xserves, however, boot by default
into the 64-bit kernel, so this is not an issue. (note that for
upcoming AT 1.3, 64-bit will be built by default for Macs boasting a
64-bit intel processor with Snow Leopard)

Regarding the best choice of MPI... I use the mpich2 variant for my
main machine (Linux box with 4 compute cores), but for my Macbook Pro,
I use OpenMPI, as that is most often suggested for Mac OS X across the
web. In fact, OpenMPI is the variant that comes pre-installed on Mac
OS X, but the fortran compiler wrappers are not configured, so it has
to be reinstalled to work for Amber anyway (just make sure that if you
want to use your own version that you install, you don't use the
existing binaries by accident).

Complaints about incompatible architectures are quite abundant I
believe for many pieces of software that you try to compile from
source (programs I've had trouble with include povray), but these
configure scripts are tricked into thinking every system library is
32-bit, so it creates 32-bit binaries to link to them, which
subsequently fails. A lot of this can be avoided by booting to the
64-bit kernel, but (at least the new) AmberTools/Amber should be able
to deal with this just fine, as there has been much discussion about
it amongst the developers.

Good luck!

Please note that the entire discussion above is about Snow Leopard, as
Leopard and earlier only have 32-bit everything.

On Tue, Dec 15, 2009 at 5:26 AM, Alan <> wrote:
> Hi David W! I always follow anything you wright about Amber on Mac and best
> Xmas gift would Ambertools 1.3. And thanks for your offlist emails with
> tips.
> I didn't check today, but did Intel finally release its compilers to work
> with Snow Leopard? There were serious issues about it.
> Your recipe AFAIR is for 32 bits only right?
> And do you still see this with intel MKL (excerpt from another email):
> "I got the absolute values identical to
> except for the sign where with MKL they were inverted (like a matrix
> multiplied by -1), e.g.:
> 537c537
> <     0.00237   -0.00485    0.00133    0.00259   -0.00563    0.00178
>  0.00236 (save)
>>    -0.00237    0.00485   -0.00133   -0.00259    0.00563   -0.00178
> -0.00236 (mine)"
> Cheers,
> Alan
> On Mon, Dec 14, 2009 at 20:01, David Watson <> wrote:
>> On Dec 14, 2009, at 12:59 PM, Freedman, Tanya wrote:
>> > Hi,
>> > I have used amber before, but I'm a beginner when it comes to installing
>> this type of package. I haven't yet been able to find very simple,
>> step-by-step guidelines for getting Amber installed and running on MacOSX.
>> Can anyone give me these instructions in the simplest, most cookbook terms
>> possible?
>> > Thanks for your help!
>> >
>> 1) Get XCode 3 from Apple or install an older version from your OS
>> installation media
>> 2) Get a fortran compiler and install it. I would recommend gfortran for
>> price, intel fortran for speed.
>>    gfortran:
>>    or intel:
>> may find it at other vendors)
>>    or absoft:
>> 3) unpack your amber tarball and learn how to use "sudo" from the
>> command-line, as you will likely want to install Amber to /usr/local/amber
>> or some such directory where you will need root permissions.
>> 4) Download the latest and greatest AmberTools (you may want to wait a week
>> or two) and install that package using the installation instructions. You
>> don't need to use MKL or intel fortran for these, and it's probably best to
>> avoid them if at all possible. Make sure that you export AMBERHOME
>> appropriately (read the instructions!) in your .bash_profile, then log out
>> and back into a Terminal session.
>> 5) Follow the instructions to build Amber in serial, but go ahead and use
>> MKL if you have the intel compilers.
>> You will have to edit your .bash_profile to export MKL_HOME. Also, make
>> note of
>> as that note is your best friend.
>> 6) If you want to use all cores on a modern system, then you will want to
>> get a working installation of mpi.
>> I would recommend mpich2, but you are going to have to download the source
>> and compile it to support your fortran compiler. Install this to
>> /opt/mpich2, then set up your .bash_profile to export MPI_HOME=/opt/mpich2
>> before you attempt to install parallel. Please read the Installer guide and
>> User guide for mpich2 if you run into problems with this one. And before you
>> attempt to compile amber parallel, be sure that you can run
>>    /opt/mpich2/bin/mpirun -np 2 hostname
>> without errors (e.g. you will see your computer's name printed on two
>> separate lines.)
>> 7) If you've gotten to this point, you may want to compile pmemd, but this
>> requires mojo and I'm not going to explain it in anymore detail unless you
>> get to this point and write me back personally.
>> Some important notes:
>> a) You will want to be familiar with the bash shell, so pick up a book on
>> that if you aren't a console guru.
>> b) Your choice of fortran compiler can make simulations go faster, but
>> newer intel compilers with MKL can cause weird problems on a few of the
>> tests.
>> c) At no point in time should you fail to read the instructions. Memorize
>> them, recite them in prayer to your deity (if applicable), and if you are
>> persistent, you will be rewarded.
>> I hope that's enough to get you pointed in the right direction.
>> I have configuration files available upon request for compiling 64 bit
>> versions of sander (serial and parallel) and pmemd with intel fortran.
>> I can't provide support for them, should you wish to use them, because you
>> will likely have a different MKL path than I do.
>> _______________________________________________
>> AMBER mailing list
> --
> Alan Wilter Sousa da Silva, D.Sc.
> PDBe group, PiMS project
> EMBL - EBI, Wellcome Trust Genome Campus, Hinxton, Cambridge CB10 1SD, UK
> +44 (0)1223 492 583 (office)
> _______________________________________________
> AMBER mailing list

Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
AMBER mailing list
Received on Tue Dec 15 2009 - 04:00:02 PST
Custom Search