Personal tools

Amber

Installing the Amber simulation package (Amber 12)

You have to purchase the Amber package -- see http://amber.scripps.edu/

See also: $AMBERHOME/INSTALL

A more complete description of the Amber 12 installation is given here: http://jswails.wikidot.com/installing-amber12-and-ambertools-12

-- under construction --

0. Install Prerequisites

  • You need a Fortran and C compiler, X11 development files and MPI development files. Amber also requires csh (or tcsh) and the patch tool for automatic updates. The GNU compilers are good starting options but commercial or otherwise optimized compilers can considerably speed up your eventual simulations.

    See: Installing Compilers (probably outdated by now)

  • the following commands should get you up and running on Ubuntu:

    sudo apt-get install gfortran gcc g++ flex patch tcsh bison libbz2-dev
    sudo apt-get install xorg-dev
    sudo apt-get install mpi-default-dev mpi-default-bin
    
  • Obsolete requirement?: Ensure you have a yacc implementation installed (try 'yacc' at the command prompt).

1. unpack the AMBER and AMBER tools package:

cd /usr/local/lib
tar xvfj Amber12.tar.bz2
tar xvfj AmberTools12.tar.bz
export AMBERHOME=`pwd`/amber12

Note:

execute the above and following commands within a root shell. For example, by first calling 'sudo zsh' or 'sudo bash'. Otherwise it is going to be more difficult to define $AMBERHOME such that the sudo command can "see" it.

2. Apply bugfixes:

  cd $AMBERHOME
  ./patch_amber.py --check-updates
  ./patch_amber.py --update-tree
  ./patch_amber.py --update-tree

Repeat the last command until now further patches are found (this is owing to updates to the patch tool itself,
which are blocking more recent updates).

3. single and MPI compilation:

GNU Fortran Compiler, single and MPI:

cd $AMBERHOME
./configure gnu
make install
./configure -mpi gnu
make install

4. Compilation for CUDA GPU computing on NVidia:

  1. download CUDA tools from NVIDIA and then:

    chmod +x cudatoolkit_4.2.9_linux_64_ubuntu11.04.run
    ./cudatoolkit_4.2.9_linux_64_ubuntu11.04.run
    
  2. create file 'nvidia_cudatools.conf' in /etc/ld.so.conf.d:

    # CUDA tools libraries
    /usr/local/cuda/lib64
    /usr/local/cuda/lib
    

    further information at: http://developer.download.nvidia.com/compute/DevZone/docs/html/C/doc/CUDA_Getting_Started_Linux.pdf

  3. Compile Amber for CUDA:

    cd $AMBERHOME
    export CUDA_HOME=/usr/local/cuda
    ./configure -cuda gnu
    make install
    

    Troubleshooting:

    This didn't work for me on Ubuntu 12.04. Compilation broke with:

    gputypes.cpp:(.text+0x1aa5): undefined reference to `cufftDestroy'
    

    and many more errors. The following hint fixed the problem: http://archive.ambermd.org/201205/0847.html Change the PMEMD_CU_LIBS variable in config.h -- put the "./cuda/cuda.a" entry at the beginning of that list:

    PMEMD_CU_LIBS=./cuda/cuda.a -L$(CUDA_HOME)/lib64 -L$(CUDA_HOME)/lib -lcurand -lcufft -lcudart
    
  4. Compile Amber for CUDA + MPI:

    cd $AMBERHOME
    export CUDA_HOME=/usr/local/cuda
    ./configure -mpi -cuda gnu
    make install
    

    Troubleshooting:

    Again this didn't work for me on Ubuntu 12.04. Compilation broke with the compiler complaing that it could not find mpi.h. The following message provided help: http://archive.ambermd.org/201206/0030.html:

    sudo apt-get install libmpich2-dev mpich2
    

    Then edit the variable PMEMD_CU_INCLUDES in config.h and add the item:

    -I/usr/include/mpich2
    

    Note: the previous compilations relied on openMPI rather than MPICH. Both systems should work. The open-mpi requirements are:

    sudo apt-get install openmpi-bin libopenmpi-dev
    

5. Test AmberSuite:

cd $AMBERHOME/test
make test

6. Permanently adapt environment (bash / zsh example):

System Message: WARNING/2 (<string>, line 126)

Literal block expected; none found.

Add these statements to your shell startup script (.bashrc, .zshenv, etc.):

export AMBERHOME=/usr/local/lib/amber12
export PATH=$PATH:$AMBERHOME/bin
export CUDA_HOME=/usr/local/cuda

Usage:
see Biskit.Amber...

Configuration: