GROMACS for Debian ------------------ The basic 'gromacs' package contains only single-threaded binaries. For multiple threads or machines, you will need the MPI-enabled version of mdrun. Depending on your MPI implementation of choice, install either the 'gromacs-openmpi' or 'gromacs-mpich'. If you are setting up an MPI installation for the first time and have no existing preference, then 'gromacs-openmpi' is your best choice. OpenMPI is currently the most active MPI development project, and has incorporated several other MPI implementations under its umbrella. Also consider installing the 'openmpi-doc' package for additional documentation. OpenMPI has superseded the LAM-MPI project. 'gromacs-mpich' supports the MPICH3 implementation. Although less commonly used in the GROMACS community than OpenMPI (and its predecessor LAM), it does support more architectures and also has an active developer base. Both of the MPI packages install their version of 'mdrun' as 'mdrun_mpi', so you can keep both regular and MPI binaries on the same machine. See the README files in /usr/share/doc/gromacs-openmpi or /usr/share/doc/gromacs-mpich (when installed) for additional information. Finally, the MPI packages can be installed independently of the main gromacs package, and unlike the main package, do not require the X Window System to be installed. This makes them suitable for installation on a large cluster, where you only intend to do serious number-crunching on that node and perform all interactive tasks on another machine. Additional Documentation ------------------------ In addition to the documentation included in this package, the GROMACS website contains several additional manuals, tutorials, and checklists of use to all skill levels, at . demux.pl and xplor2gmx.pl ------------------------- You may see references to these scripts on the GROMACS mailing lists. In this package, they have been renamed without their suffixes: demux and xplor2gmx. Double Precision Binaries ------------------------- The standard version of all binaries have been compiled for single- precision floating point operations. For some applications, this may result in unacceptably large numeric instabilities. Double-precision calculations can often solve such problems, with a tradeoff in speed (typically a ~30% slowdown on i386). Binaries compiled for DP have a "_d" suffix, and in all other ways operate identically to their single- precision counterparts. CPU Optimizations ----------------- On amd64 and i386 architectures, the compiler is restricted to SSE2 extensions. Some systems may have additional SIMD extensions available that can speed execution by large amounts, but are not available on all CPUs of that architecture (e.g. AVX extensions on amd64 that are only present on Intel's Sandy Bridge or AMD Bulldozer and subsequent chips). To compile a package with autodetection of available extensions, use the environment variable DEB_BUILD_OPTIONS=cpuopt . This choice of default should work on all amd64 CPUs, and i386 CPUs from the Intel Pentium 4 or AMD Opteron/Athlon 64 and forward (roughly speaking, all new systems since 2001). Although this does exclude certain very old i386 systems, the performance tradeoffs from disabling SSE2 are severe; you may not get much performance-per-watt on such CPUs. On such CPUs, compiling your own binaries is mandatory. For more information about GROMACS and CPU acceleration, please see . -- Nicholas Breen , Fri, 15 Nov 2013 11:12:19 -0800