GROMACS 4.5.1 is bug fix release.
Release notes for 4.5
New features
* Pencil decomposition of the reciprocal space PME grid to
improve scaling. This reduces the amount of communication
for high parallelization and improves load balancing with up
to 40% overall performance improvement for large systems.
* Memory usage is improved for very large systems, allowing
simulations of >100 million atoms.
* Running on a multi-core node now uses thread-based
parallelization to automatically spawn the optimum number of
threads in the default build. MPI is now only required for
parallelization over the network.
* Domain decomposition can now also be used without periodic
boundary conditions
* GPU acceleration support on NVIDIA cards. This first release
with GPU support based on OpenMM provides up to an order of
magnitude faster performance for implicit solvent simulations,
but PME simulations are about as fast as on a high-end CPU.
* Check-pointing is made more secure:MD5sum are used to verify
that all files are correctly in-place before a simulation is
appended. Output file appending at continuation is turned on
by default.
* Increased tolerance for networked file system failures and
cluster node crashes: checkpoint handling is safer and mdrun
forces file system cache flushes during checkpoints.
* Full CMake support. After the 4.5 release we will be
switching the default build tool from autoconf to cmake,
and possibly deprecate autoconf in the future.
* Full support for seven AMBER force fields in the standard
distribution, with default Amber names. We also include the
recent Amber99sb-ildn in the distribution.
* Support for CHARMM27, including cmap for dihedrals
* Efficient Generalized-Born implicit solvent support
including the Still/HCT/OBC-models to compute the Born radii,
a novel way of tabulating the generalized Born-interaction
formula for greater speed, and optimized SSE-routines in both
single and double precision.
* Highly efficient all-vs-all assembly kernels for both vanilla
and generalized born interactions, in both single and double
precision.
* Much better support for nucleic acid simulations, including
automatic handling by pdb2gmx.
* Support for Velocity-Verlet integrators for reversible T-
and P-coupling; MTTK pressure control integrators;
Nose-Hoover chains.
* Symplectic Trotter Leap-Frog integrator for twin-range
non-bonded interactions.
* Support for Bennet acceptance ratio calculations through
direct calculation of Hamiltonian differences during the
simulation.
* File formats: All GROMACS tools can now read any VMD
supported trajectory format, without converting trajectory
first. (VMD libraries are required).
* pdb2gmx now retains the residue numbers from the input,
mdrun and all tools use these original numbers.
New tools
* g_bar: Bennett acceptance ratio (BAR) free energy calculations,
including automatic error estimates and phase space overlap
measures.
* g_rdf was a little bit enhanced that structure factors can
be calculated for any system, by supplying the necessary data
via sfactor.dat. Most of the common atomtypes are already
contained, but everybody who needs more freedom can enhance
the table
* g_select: Library support for "dynamic index groups" based
on textual selections (experimental feature).
See the tool g_select, the included template.c, or Doxygen
documentation for information on how to write analysis tools
using the library. Existing tools have not (yet) been
converted.
* g_tune_pme: For a given number of processes or threads this
tool systematically times mdrun with various numbers of
PME-only nodes and determines which setting is fastest. It
also checks whether performance can be enhanced by shifting
load between the real and the reciprocal space part of the
Ewald sum.
* g_membed: a very convenient utility for rapidly embedding
membrane proteins into equilibrated lipid bilayers
* g_pme_error: estimates the error of the electrostatic forces
if using the SPME algorithm. TO be incorporated in g_tune_pme
Changes that might affect your results
* grompp by default sets the new nstcalcenergy parameter equal
to nstlist, this has no effect on the integration, only on
the energy averages stored in ener.edr
* grompp by default sets the new nsttcouple parameter equal to
nstlist, this means T-coupling is done less frequently;
grompp checks if tau_t is large enough
* grompp by default sets the new nstpcouple parameter equal to
nstlist, this means P-coupling is done less frequently;
grompp checks if tau_p is large enough
* mdrun results with old tpr files with twin-range non-bonded
interactions will be different, because of the new symplectic
integrator
* for free-energy calculations sc-sigma now also sets the minimum
soft-core sigma (old tpr files retain the old behavior,
which can be enforced by setting the env.var. GMX_SCSIGMA_MIN to 0)
e46ce0136a