GROMACS 5.1 Release NotesΒΆ
These release notes document the functionality changes that have taken place in GROMACS since version 5.0.
Some bug fixes are mentioned here, but those fixed in 4.6 or 5.0 branches are (or will be) documented there. Such fixes can generally be assumed to be present in patch releases of subsequent major/minor versions.
- Improvements to mdrun performance
- Improved default threading configuration
- Made it harder to use slow parallelism setups
- Improved performance auto-tuning with GPUs
- Added kernel and compiler suport for latest CUDA CC 3.7 devices
- Added optional NVIDIA Management Library (NVML) Integration
- Allowed increasing CUDA thread block size
- Added CUDA compiler support for CC 5.0
- Added support for flushing WDDM queue on Windows
- Optimized atomic accumulation in CUDA short-ranged kernels
- Improved pair search thread load balance
- Improved the intra-GPU load balancing
- Added support for arbitrary number of OpenMP threads
- Improved performance of bonded interactions from use of SIMD
- Added checks for inefficient resource usage
- Made it possible to use 1 PP and 1 PME rank
- Reduced the cost of communication in the pull code
- Added SIMD acceleration for LINCS
- Improved OpenMP scaling of LINCS
- Improved SIMD support in LINCS and bonded interactions
- Cleaned up unused code paths that made some kinds of calculations slow
- Fixed for processors being offline on Arm
- New features
- Added support for OpenCL acceleration
- Added support for inter-molecular bonded interactions
- Added flat-bottomed potential to pull-coordinate types
- Allowed pull groups of 1 atom with mass 0
- Changed the way pull directions are selected
- Improved pull geometry cylinder
- Expanded pulling output options
- Allowed configure-time specification target GPU architectures
- Added single-accuracy SIMD double math functions
- Added new tests
- Improved parallel distribution and thread count reporting
- Implemented zsh shell completions for gmx command
- New fatal error to stop benchmarking runs if tuning is still active
- Changes to GROMACS tools
- Improved selection error messages (use exceptions)
- Supported more complex fixed position selections
- Fixed crash with ‘z of ...’ within a selection expression
- Reworked
gmx solvate,gmx insert-moleculesandgmx sasa - Converted
gmx rdfto the C++ analysis framework - Added basic GPU support for gmx tune-pme
- Required
gmx tune_pmeto take -mdrun parameter - Added
gmx pairdisttool as partialgmx mindistreplacement - Fixed
gmx trjconvmodifying frame time stamp unnecessarily - Reorganized analysis of and documenting correlation functions
- Made ‘gmx sasa’ calculation thread-safe
- Made ‘gmx sasa’ volume computation translation-invariant
- Added OpenMP parallelization to
gmx wham
- New and reorganized documentation
- Features removed in this version
- Removed temporary symlinks from pre-5.0 tool names
- Removed support for partial matches for selection keywords
- Removed
mdrun -testverlet - Removed iteration + constraints framework
- Removed heuristic group-scheme neighbour-list updates
- Removed
mdrun -seppot - Removed support for
.tpa, .tpb, .tpx, .trjfile formats - Removed
gmx protonatetool
- Miscellaneous changes
- Updated physical constants from NIST 2010 CODATA
- Hardware reporting now covers all nodes
- Avoided near-silent generation of zero energies during reruns
- Added tip4pew parameters to oplsaa.ff
- Improved and clarify mdrun behavior with nsteps<0
- Updated bundled TNG version to 1.7.6
- Handled integer vectors correctly in DD error message
- Added support for MemorySanitizer checking
- Enabled fp-exceptions
- Fixed constraints checks in grompp
- Made mdrun print the list of compatible GPUs
- Added .travis.yml
- Added flat-bottom cylindrical restraints in x and y.
- Updated template/README
- Used native CMake mechanism for find_package(GROMACS)
- Added input file options that accept missing files
- Fixed double precision reference SIMD and gcc bug
- Changed
mdrun -deffnmto append_pullx_and_pullfto pull files - Corrected nstcalclr/nstfep/nst_repl_ex checks
- Fixed compile issues on K computer
- Reenabled SIMD kernels when running without cut-offs
- Changed to use MPI_THREAD_FUNNELED when available
- Replaced hardware threads by logical cores in output
- Updated FFTW recommendations in install-guide.
- Automated the
mdrun -gpu_idoption - Updated
gmxhelp output - Removed non-standard atom types from OPLS/AA and CHARMM27
- Made bond-atomtypes case sensitive
- Removed information about obsolete .edo file format
- Fixed PGI compiler flag
- Added more new quotes
- New low-level SIMD support