Versions 4.6.x

    Version as of 11:24, 1 Aug 2021

    to this version.

    Return to Version archive.

    View current version

    Release notes for 4.6.5

    • Fix GPU-load balancing and GPU-sharing bugs introduced in a bug fix in 4.6.4 #1385
    • Minor fix to possible Verlet-scheme memory issue with SIMD on 32-bit builds
    • Minor fix to variable name in documentation
    • Fix harmless bug with useless sorting of bonded interactions in free-energy\ calculations #1387

    Release notes for 4.6.4

    • Implemented plain-C SIMD macros for testing, reference, and development #1173
    • Introduced general 4-wide SIMD support for PME spread+gather and the NBNxN searching
    • Minor optimization to generic SIMD invsqrt #1333
    • Fixed bug with the SIMD padding in the case of very long group-scheme neighbor lists #1341.
    • Fixed rare division by zero in SIMD angles and dihedrals #1351
    • Fixed the NBNxN buffer size estimate to be more correct in the presence of vsites, and be less conservative overall. This enhances performance, particularly with GPUs.
    • Consolidated and refactored NBNxN SIMD kernel utility routines, and bounding box data structures. Split up Verlet SIMD kernels for faster compilation.
    • Added BlueGene/Q Verlet cut-off scheme kernels, enhancements to CMake handling, support for bgclang (but latest compiler does not yet work with OpenMP), support for A2 core and QPX SIMD in CPU detection, updates to install guide.
    • Removed (harmless) left-over in NBNxN SIMD kernels, improving performance of PME + pressure coupling by about 5%.
    • Fixed atom sorting with NBNxN kernels with bonded interactions at long ranges
    • Reorganized GPU detection, and selection. Clarified reporting on and documentation of GPU usage.
    • Fixed GPU detection to occur only once per physical node #1358
    • Enabled GPU sharing among tMPI ranks
    • Corrected dynamic load balancing when MPI ranks share GPUs
    • Added support for using CUDA texture objects, using CUDA stream priorities, and compilation that can optimize for CUDA compute capabaility 3.5
    • Fixed tMPI_Atomic_memory_barrier for Xeon Phi
    • Enhanced configure-time support for testing for the presence of atomic operations. #1355
    • Fixed issue where OpenMP threads could be pinned to the same cores #1360
    • Fixed multiple distance restraints with OpenMP #1316
    • mdrun without OpenMP with thread-MPI now uses all cores #1317
    • Corrected volume with serial NPT replica exchange #1362
    • Fixed restart from checkpoint re-initializin Wang-Landau weights in expanded ensemble calculations. #1350
    • Disabled expanded ensemble for all integrators but md-vv #1321
    • Fixed problem caused by numerical overflow in expanded ensemble #1314
    • Fixed a problem with type 2 pair interactions in free-energy calculations #1315
    • Fixed reaction-field free energy bug #1318
    • Fixed logic for free energies with mdrun -rerun #1330
    • Fixed sc-coul description in manual #1331
    • Made free energy PME kernel 40% faster
    • Fixed minor admin issues (pkg-config, install guide, date and version stamps, removing intermediate and output files from repo, cmake warning and error messages, cmake functionality, ctest usage, compiler warnings, fixing typos)
    • Fixed possible dereference of null pointer on compute nodes that lack getpwuid() #1301
    • Fixed pdb2gmx -vsite hydrogen -o conf.pdb
    • Removed buggy -seppot output #1294
    • Fixed a half bin misalignment in gmx_vanhove -or
    • Fixed limitations in number of frames g_cluster can handle
    • Various fixes to g_tune_pme (no assumption that MPI is available, better error messages) #1319
    • Fixed reading history_t from checkpoint #1174
    • Fixed total time measurement with separate PME nodes #1325
    • Fixed mdrun -version to finalize MPI properly #1313
    • Fixed parallel normal modes with PME. Also fixed some EM/NM output layout. #1308
    • Fixed GMX_DD_DUMP_GRID output
    • Fixed minor wallcycle output issues
    • Corrected grompp constraint/DOF warning with vsites #1322
    • Fixed uninitialized error in g_enemat #1312
    • Updated configure-time management of linear algebra libraries #771,#1186
    • Enabled finding mkl.h on more icc versions #1110
    • Updated dlist.c to help analysis tools like g_chi recognize more atom names.

    Release notes for 4.6.3

    • Fixed concurrency issues with thread affinity settings with or without MPI #1270, #1254. Note that #1254 issue 3 seems to be an OpenMPI bug
    • Split cmake thread detection into core thread and thread_mpi
    • Fixed deadlock on Mac OS X during thread affinity check
    • Fixed inconsistent locality_order array for thread affinity
    • Fixed issues with code for atomic operations with XLC and on K-computer #1284, #1274
    • Really fixed SD and BD integrator OpenMP performance #1121
    • Fixed deadlock with replica exchange and -maxh #1273
    • Fixed DD internal state corruption during energy minimization #1272
    • Fixed syntax for FAHCORE in CMakeLists.txt
    • Fixed flops table to report CUDA analytical Ewald contribution
    • Fixing a problem when dhdl and replex are not multiples
    • Added CUDA compiler flags to version header
    • Fixed mdrun build to work with ORCA #1286
    • Corrected definition of EEL_USER #1289
    • Removed genion non-random insertion of ions, also stopped writing useless log file, and added warning about -conc option over-riding -nn or -np. #1236, #615, and #1208
    • Fixed minor g_bar output issues
    • Fixed -npstring command line argument in g_tune_pme -npstring
    • Added GMX_NO_CREDITS environment variable to disable showing credits. #1267
    • Removed unused variables
    • Fixed a CMake typo #1280
    • Fixed various CMake and compiler issues for BlueGene/Q #1281, #1282, #1283
    • Minor tweaks to qm_orca.c
    • Fixed some inconsistencies in pkg-config files

    Release notes for 4.6.2

    • Fixed linking with FFTW when using MKL for BLAS & LAPACK #1067
    • Made FFT detection more quiet when nothing was changing
    • Updated mechanism and documentation for linking to MKL for FFT, BLAS and LAPACK #1110 #1186
    • Added compiler flags to suppress over-zealous warnings from gcc 4.8
    • Added workaround for clang compiler on AMD FMA processors #1099
    • Fixed automated download of correct regression tests when working with git version
    • Added install guide section for BLAS & LAPACK #1186
    • Corrected grompp rvdw charge-group radii check #1164
    • Updated grompp cut-off and PME-Switch checks if the switch distance is too large #1179
    • Allowed PME to work with different numbers of OpenMP threads available on different MPI ranks #1171
    • Fixed PME time printing with -ntomp not equal to -ntomp_pme #1158
    • Improved PME load balancing by checking for grid restrictions
    • Improved load balancing with Verlet cut-off scheme
    • Fixed mdrun walltime reporting in .log file #1210
    • Added acceleration paths suitable for Fujitsu Sparc64, particularly group kernels for the K computer.
    • Added SIMD acceleration for angles and dihedrals
    • Refactored NxN kernels to use generic SIMD operations
    • Added CUDA PME kernels with analytical Ewald correction
    • Fixed mdrun -nsteps to handle large numbers of steps #1224
    • Fixed v-rescale thermostat to work with tau-t >= 0
    • Fixed copious issues with free-energy calculations, including 1-4 interactions #1225, Urey-Bradley angles in CHARMM27 #1115, perturbations of mass #1232, reruns from non-trr input #1240, bond constraints #1255
    • Added work-around for no-PBC, infinite-cutoff so that these work correctly (but are slower than 4.5-era code) #1249 #1095
    • Introduced fatal error for generalized Born free-energy calculations, since these do not work #1237
    • Fixed memory leak in expanded ensemble code #1265, and other fixes
    • Fixed reading old .tpr files with dihedral restraints #1194
    • Fixed enforced rotation with Verlet cut-off scheme #1155
    • Fixed g_hydorder and PBC #1238
    • Fixed incorrect scaling of cos-acceleration viscosity #1244
    • Improvements to OpenMM build (but still broken)
    • Fixed bugs in g_select subexpression handling #1216 #1219
    • Fixed g_sgangle legends
    • Fixed legend in essential dynamics output file
    • Fixed bug with g_mindist #1183
    • Fixed g_chi -omega to follow IUPAC definitions of dihedrals #953
    • Clarified g_gyrate -h #934
    • Fixed error in g_select online help #1262
    • Fixed -neutral option in genion, and -nq and -pq with ions of charge > 1 when using -neutral
    • Added GROMOS96 54A7 files from ATB website #773
    • Many minor enhancements to ThreadMPI
    • Fixed potential race conditions in ThreadMPI thread creation with pthreads #1254
    • Fixed thread offset & stride check
    • Improved thread-pinning informational messages
    • Fixed a thread-safety issue in affinity layout detection #1254
    • Improved logic about whether to switch to polling GPU wait
    • Fixed several remaining reference to outdated license

    Release notes for 4.6.1

    • increased shared object major version to 8 (should have done this for 4.6, sorry) #1147
    • updates to HTML manual, install guide, PDF manual, shell completions
    • copious minor bug fixes
    • various build system upgrades and fixes #1143
    • new and enhanced error messages
    • fixes for AdResS bugs (neighbour list construction, flop accounting, multiple tf tables)
    • fixed PME timing counter issues #1125
    • fixed PME load balance reporting
    • fixed forcerec to work with tools like genion and g_disre #1136
    • various GPU performance enhancements
    • fixed sd integrator with OpenMP threading #1138
    • various minor fixes for interacting with CUDA for GPUs
    • fixes for g_tune_pme to cope with new mdrun behaviour and changed command-line options (for both g_tune_pme and mdrun)
    • more checks for system support for setting thread affinities
    • removed inter-flag dependency in g_order
    • fixed issues with free-energy pertubation soft-core and cut-offs #1146
    • fixed issues with md-vv + nose-hoover + (nstcalcenergy > nsttcouple) #1129
    • incorporated new changes from release 4.5.x branch
    • prevented building with icc 11.1 and SSE4.1 because of known problems #1126
    • adding warning about not building with icc version < 12 #1126
    • fixed bug sorting atoms with GPUs introduced since 4.6 #1153
    • fixed issues with automated download of regression tests #1150
    • fixed bug with DD cut-off check and PME dynamic load balancing #1169

    Release notes for 4.6 (2013-01-19)

    New features
    • New Verlet non-bonded scheme which, by default, uses exact cut-off's and a buffered pair-list.
    • Multi-level hybrid parallelization (MPI + OpenMP + CUDA):
      • full OpenMP multithreading with the Verlet scheme;
      • OpenMP mulitthreading for PME-only nodes with the group scheme;
      • native GPU acceleration using CUDA (supporte NVIDIA hardware).
    • New x86 SIMD non-bonded kernels for the usual cut-off scheme, called group scheme and the new verlet scheme, use x86 SIMD intrinsics (no more assembly code):
      • SSE2
      • SSE4.1
      • AVX-128-FMA (for AMD Bulldozer/Piledriver)
      • AVX-256 (for Intel Sandy/Ivy Bridge)
    • Improved PME spread, gather, solve and FFT communication, total improvement ~25% on x86.
    • Automated OpenMP thread count choice to use all available cores.
    • Automated CPU affinity setting: locking processes or threads to cores.
    • Automated PP-PME (task) load-balancing: balancing non-bonded force and PME mesh workload when the two are executed on different compute-resources (i.e CPU and GPU or different CPUs). This enables GPU-CPU and PP-PME process load balancing by shifting work from the mesh to the non-bonded calculation.
    • PPPM/P3M with analytical derivative at the same cost and with the same features as PME.
    • New, advanced free energy sampling techniques.
    • AdResS adaptive resolution simulation support.
    • Enforced rotation ("rotational pulling")
    • Build configuration now uses CMake, configure+autoconf/make no longer supported. (The CMake build system features with a lot of automation and cleverness under the hood and we know that the it might not always prove to be as rock-solid as the old one. However, far more advanced and complex, so bear with us while we iron out issues that come up along the way.)
    • Improved regressiontests; these can now be run directly from the build tree using make check
    • g_hbond now utilizes OpenMP.

    No critical bugfixes. This version is based on 4.5.6 and all important fixes are "inherited"  and therefore documented in the 4.5.6 release notes.

    Changes that might affect your results

    None for simulations set up with the traditional group cut-off scheme.

    When switching from the group scheme to the Verlet scheme, integration of the equations of motion can get more accurate due to the exact cut-off treatment and buffering (this will, of course, depend on the original cut-off settings used). See the section Cut-off schemes for details.

    Other important changes compared to 4.5
    mdrun does now thread affinity setting
    This means that when runing multiple mdrun processes on the same machine, one has to either provide a core "pin offset" using the -pinoffset command line option, or turn off internal affinities and take the performance hit (or alternatively manage affinities externally).
    The choice of compiler matters more
    With the switch to SIMD intrinsics, up-to-date SIMD CPU acceleration support, OpenMP, the compiler used matters more both in terms the ability to compile GROMACS correctly  and from the point of view of mdrun performance. The recommended compilers that are known to work (=compile GROMACS correctly) and provide good performance on x86/AMD64 are: gcc 4.5 and later, Intel Compilers 12.0 and later and clang 3.1 (note the lack of OpenMP support which can cause 30%+ performance loss). In all cases you are strongly advised to use the most recent patch level available. GROMACS makes extensive use of compiler intrinsics to get the most out of your hardware, so if you use a compiler that is older than your hardware you are asking for trouble, because all the compilers have had bugs in their intrinsics implementations. For further details see ???.
    Bugfixes and improvements since beta3
    • fixed performance bug with SD+BD integration and OpenMP multi-threading (#1121)
    • fixes to free energy code, output & g_bar compatibility (#1090)
    • fixed multi-threading with hybrid GPU+CPU mode (#1100)
    • fixed GB interactions (#1096)
    • fixed issues with pressure control and infrequent evaluation
    • fixes for md-vv and rerun
    • fixed resetting states with parallel Verlet scheme
    • fixed nbnxn no LJ comb.rule AVX256 PME kernel
    • fix for compiler flag handling (#1038, #1040)
    • fixed bug with Verlet + DD + bonded atom communication
    • fixed SSE/AVX compilation under Windows (#1092, #1093, #1068)
    • fixed a bug with multiple exchanges
    • fixed nbnxn AVX-256 Ewald table pointer alignment
    • fixed GMXRC so we are not polluting standard shell variables
    • thread-MPI fixes for i386 llvm & simplification of atomics
    • added work-around for gcc bug in AVX intrinsincs formal parameter


    • regressiontests can be run from the build tree now (make check)
    • efficiency improvements for PP-PME load balancing + DD DLB
    • using topology information for thread affinity setting
    • added mdp option 'calc-lambda-neighbors'
    • made g_tune_pme work correctly again with thread-MPI mdrun
    • Blue Gene build system support and documnetation
    • made SHAKE work again with particle decompostion
    • g_sans - add trajectory avereging


    4.6-beta3 (2012-12-22)

    Bugfixes and improvements
    • fixed pressure in MTTK when using constrants and dispersion correction (#1061)
    • fixed expanded ensemble and Hamiltonian replica exchange
    • fixed Andersen temperature coupling use of random number generator in parallel
    • fixed Andersen temperature coupling when constraints present
    • fixed LINCS with virtual sites - there were bugs with how LINCS did constraining based upon the forces introduced when some parallelization was added previously
    • forced c++ linking of GPU utility routines
    • fixed GPU pair search when not using x86 CPU acceleration (#1042, #1062)
    • fixed compilation with PGI compiler
    • fixes to pull code (#1071)
    • fixed license details reported by GROMACS tools
    • completed removal of Fortran kernels (we are not aware of any systems where these would run faster than the corresponding non-accelerated C kernels by enough to be worth our effort, and probably the new force-only C kernels will be faster than the old Fortran kernels on any system where the disparity between Fortran and C compiler optimization is noticeable; speak up if any of this is a problem for you!)
    • completed removal of Power6 accelerated kernels (currently we lack the resources to implement accelerated kernels for Power architectures, and probably the new force-only C kernels will show results comparable with the old accelerated Power kernels; speak up if any of this is a problem for you - particularly if you have resources to offer to fix it!)
    • re-implemented Verlet kernels on AVX-256 hardware for better performance
    • prepared Verlet kernels for future development of non-x86 SIMD support (e.g. BlueGene SIMD acceleration support is planned)
    • improvements to functioning and documentation of mechanism to specify GPU IDs to mdrun 
    • improved communication when using P-LINCS and only constraints on bonds to hydrogen
    • GROMACS provides template code for user implementations of a custom GROMACS tool in share/template, which now builds by default (and correctly!), but does not install
    • fixed header file for use by external code linking to GROMACS
    • fixes to the Reference CMake build type we used for generating reference versions of our regression tests
    • added CMake option to disable printing of GROMACS cool quotes
    • minor improvements to CMake messages to user
    • CMake cleanup

    4.6-beta2 (2012-12-06)

    Bugfixes and improvements
    • re-enabled AdResS feature (only generic kernels for now);
    • improved OpenMP parallelization performance of non-bonded force calculation with Verlet scheme;
    • fixed segv in Verlet pair-search with trilinic domain-decomposition;
    • fixed incorrect virial with virtual sites and OpenMP;
    • fixed labelling of g_hbond plots;
    • fixed compilation issue with cmake 2.8.10 and GPU acceleration;
    • fixed issues with multi-sim runs and GPU-acceleration.

    4.6-beta1 (2012-11-30)

    First beta, yay*! See the release notes above.

    (*No previous version in the 4.6 series so no list of bugfixes and improvements here.)

    Page last modified 22:59, 2 Dec 2013 by mabraham