4 |Gromacs| programs may be influenced by the use of
5 environment variables. First of all, the variables set in
6 the ``GMXRC`` file are essential for running and
7 compiling |Gromacs|. Some other useful environment variables are
8 listed in the following sections. Most environment variables function
9 by being set in your shell to any non-NULL value. Specific
10 requirements are described below if other values need to be set. You
11 should consult the documentation for your shell for instructions on
12 how to set environment variables in the current shell, or in configuration
13 files for future shells. Note that requirements for exporting
14 environment variables to jobs run under batch control systems vary and
15 you should consult your local documentation for details.
20 Print constraint virial and force virial energy terms.
23 |Gromacs| automatically backs up old
24 copies of files when trying to write a new file of the same
25 name, and this variable controls the maximum number of
26 backups that will be made, default 99. If set to 0 it fails to
27 run if any output file already exists. And if set to -1 it
28 overwrites any output file without making a backup.
31 if this is explicitly set, no cool quotes
32 will be printed at the end of a program.
35 prevent dumping of step files during
36 (for example) blowing up during failure of constraint
40 dump all configurations to a :ref:`pdb`
41 file that have an interaction energy less than the value set
42 in this environment variable.
45 ``GMX_VIEW_XVG``, ``GMX_VIEW_EPS`` and ``GMX_VIEW_PDB``, commands used to
46 automatically view :ref:`xvg`, :ref:`xpm`, :ref:`eps`
47 and :ref:`pdb` file types, respectively; they default to ``xv``, ``xmgrace``,
48 ``ghostview`` and ``rasmol``. Set to empty to disable
49 automatic viewing of a particular file type. The command will
50 be forked off and run in the background at the same priority
51 as the |Gromacs| tool (which might not be what you want).
52 Be careful not to use a command which blocks the terminal
53 (e.g. ``vi``), since multiple instances might be run.
55 ``GMX_VIRIAL_TEMPERATURE``
56 print virial temperature energy term
59 the size of the buffer for file I/O. When set
60 to 0, all file I/O will be unbuffered and therefore very slow.
61 This can be handy for debugging purposes, because it ensures
62 that all files are always totally up-to-date.
65 set display color for logo in :ref:`gmx view`.
67 ``GMX_PRINT_LONGFORMAT``
68 use long float format when printing
72 Applies for computational electrophysiology setups
73 only (see reference manual). The initial structure gets dumped to
74 :ref:`pdb` file, which allows to check whether multimeric channels have
75 the correct PBC representation.
77 ``GMX_TRAJECTORY_IO_VERBOSITY``
78 Defaults to 1, which prints frame count e.g. when reading trajectory
79 files. Set to 0 for quiet operation.
81 ``GMX_ENABLE_GPU_TIMING``
82 Enables GPU timings in the log file for CUDA. Note that CUDA timings
83 are incorrect with multiple streams, as happens with domain
84 decomposition or with both non-bondeds and PME on the GPU (this is
85 also the main reason why they are not turned on by default).
87 ``GMX_DISABLE_GPU_TIMING``
88 Disables GPU timings in the log file for OpenCL.
92 ``GMX_PRINT_DEBUG_LINES``
93 when set, print debugging info on line numbers.
96 number of steps that elapse between dumping
97 the current DD to a PDB file (default 0). This only takes effect
98 during domain decomposition, so it should typically be
99 0 (never), 1 (every DD phase) or a multiple of :mdp:`nstlist`.
101 ``GMX_DD_NST_DUMP_GRID``
102 number of steps that elapse between dumping
103 the current DD grid to a PDB file (default 0). This only takes effect
104 during domain decomposition, so it should typically be
105 0 (never), 1 (every DD phase) or a multiple of :mdp:`nstlist`.
108 general debugging trigger for every domain
109 decomposition (default 0, meaning off). Currently only checks
110 global-local atom index mapping for consistency.
113 over-ride the number of DD pulses used
114 (default 0, meaning no over-ride). Normally 1 or 2.
116 ``GMX_DISABLE_ALTERNATING_GPU_WAIT``
117 disables the specialized polling wait path used to wait for the PME and nonbonded
118 GPU tasks completion to overlap to do the reduction of the resulting forces that
119 arrive first. Setting this variable switches to the generic path with fixed waiting
122 There are a number of extra environment variables like these
123 that are used in debugging - check the code!
125 Performance and Run Control
126 ---------------------------
127 ``GMX_DO_GALACTIC_DYNAMICS``
128 planetary simulations are made possible (just for fun) by setting
129 this environment variable, which allows setting :mdp:`epsilon-r` to -1 in the :ref:`mdp`
130 file. Normally, :mdp:`epsilon-r` must be greater than zero to prevent a fatal error.
131 See webpage_ for example input files for a planetary simulation.
133 ``GMX_ALLOW_CPT_MISMATCH``
134 when set, runs will not exit if the
135 ensemble set in the :ref:`tpr` file does not match that of the
138 ``GMX_CUDA_NB_EWALD_TWINCUT``
139 force the use of twin-range cutoff kernel even if :mdp:`rvdw` equals
140 :mdp:`rcoulomb` after PP-PME load balancing. The switch to twin-range kernels is automated,
141 so this variable should be used only for benchmarking.
143 ``GMX_CUDA_NB_ANA_EWALD``
144 force the use of analytical Ewald kernels. Should be used only for benchmarking.
146 ``GMX_CUDA_NB_TAB_EWALD``
147 force the use of tabulated Ewald kernels. Should be used only for benchmarking.
149 ``GMX_CUDA_STREAMSYNC``
150 force the use of cudaStreamSynchronize on ECC-enabled GPUs, which leads
151 to performance loss due to a known CUDA driver bug present in API v5.0 NVIDIA drivers (pre-30x.xx).
152 Cannot be set simultaneously with ``GMX_NO_CUDA_STREAMSYNC``.
154 ``GMX_DISABLE_CUDALAUNCH``
155 disable the use of the lower-latency cudaLaunchKernel API even when supported (CUDA >=v7.0).
156 Should only be used for benchmarking purposes.
158 ``GMX_DISABLE_CUDA_TIMING``
159 Disables GPU timing of CUDA tasks; synonymous with ``GMX_DISABLE_GPU_TIMING``.
162 times all code during runs. Incompatible with threads.
164 ``GMX_CYCLE_BARRIER``
165 calls MPI_Barrier before each cycle start/stop call.
168 build domain decomposition cells in the order
169 (z, y, x) rather than the default (x, y, z).
171 ``GMX_DD_USE_SENDRECV2``
172 during constraint and vsite communication, use a pair
173 of ``MPI_Sendrecv`` calls instead of two simultaneous non-blocking calls
174 (default 0, meaning off). Might be faster on some MPI implementations.
176 ``GMX_DLB_BASED_ON_FLOPS``
177 do domain-decomposition dynamic load balancing based on flop count rather than
178 measured time elapsed (default 0, meaning off).
179 This makes the load balancing reproducible, which can be useful for debugging purposes.
180 A value of 1 uses the flops; a value > 1 adds (value - 1)*5% of noise to the flops to increase the imbalance and the scaling.
182 ``GMX_DLB_MAX_BOX_SCALING``
183 maximum percentage box scaling permitted per domain-decomposition
184 load-balancing step (default 10)
186 ``GMX_DD_RECORD_LOAD``
187 record DD load statistics for reporting at end of the run (default 1, meaning on)
189 ``GMX_DETAILED_PERF_STATS``
190 when set, print slightly more detailed performance information
191 to the :ref:`log` file. The resulting output is the way performance summary is reported in versions
192 4.5.x and thus may be useful for anyone using scripts to parse :ref:`log` files or standard output.
194 ``GMX_DISABLE_SIMD_KERNELS``
195 disables architecture-specific SIMD-optimized (SSE2, SSE4.1, AVX, etc.)
196 non-bonded kernels thus forcing the use of plain C kernels.
198 ``GMX_DISABLE_GPU_TIMING``
199 timing of asynchronously executed GPU operations can have a
200 non-negligible overhead with short step times. Disabling timing can improve performance in these cases.
202 ``GMX_DISABLE_GPU_DETECTION``
203 when set, disables GPU detection even if :ref:`gmx mdrun` was compiled
206 ``GMX_GPU_APPLICATION_CLOCKS``
207 setting this variable to a value of "0", "ON", or "DISABLE" (case insensitive)
208 allows disabling the CUDA GPU allication clock support.
210 ``GMX_DISRE_ENSEMBLE_SIZE``
211 the number of systems for distance restraint ensemble
212 averaging. Takes an integer value.
215 emulate GPU runs by using algorithmically equivalent CPU reference code instead of
216 GPU-accelerated functions. As the CPU code is slow, it is intended to be used only for debugging purposes.
219 disable exiting upon encountering a corrupted frame in an :ref:`edr`
220 file, allowing the use of all frames up until the corruption.
223 update forces when invoking ``mdrun -rerun``.
226 set in the same way as ``mdrun -gpu_id``, ``GMX_GPU_ID``
227 allows the user to specify different GPU IDs for different ranks, which can be useful for selecting different
228 devices on different compute nodes in a cluster. Cannot be used in conjunction with ``mdrun -gpu_id``.
231 set in the same way as ``mdrun -gputasks``, ``GMX_GPUTASKS`` allows the mapping
232 of GPU tasks to GPU device IDs to be different on different ranks, if e.g. the MPI
233 runtime permits this variable to be different for different ranks. Cannot be used
234 in conjunction with ``mdrun -gputasks``. Has all the same requirements as ``mdrun -gputasks``.
236 ``GMX_IGNORE_FSYNC_FAILURE_ENV``
237 allow :ref:`gmx mdrun` to continue even if
241 when set to a floating-point value, overrides the default tolerance of
242 1e-5 for force-field floating-point parameters.
244 ``GMX_MAXCONSTRWARN``
245 if set to -1, :ref:`gmx mdrun` will
246 not exit if it produces too many LINCS warnings.
249 use the generic C kernel. Should be set if using
250 the group-based cutoff scheme and also sets ``GMX_NO_SOLV_OPT`` to be true,
251 thus disabling solvent optimizations as well.
254 neighbor list balancing parameter used when running on GPU. Sets the
255 target minimum number pair-lists in order to improve multi-processor load-balance for better
256 performance with small simulation systems. Must be set to a non-negative integer,
257 the 0 value disables list splitting.
258 The default value is optimized for supported GPUs (NVIDIA Fermi to Maxwell),
259 therefore changing it is not necessary for normal usage, but it can be useful on future architectures.
262 use neighbor list and kernels based on charge groups.
265 when set, print detailed neighbor search cycle counting.
267 ``GMX_NBNXN_EWALD_ANALYTICAL``
268 force the use of analytical Ewald non-bonded kernels,
269 mutually exclusive of ``GMX_NBNXN_EWALD_TABLE``.
271 ``GMX_NBNXN_EWALD_TABLE``
272 force the use of tabulated Ewald non-bonded kernels,
273 mutually exclusive of ``GMX_NBNXN_EWALD_ANALYTICAL``.
275 ``GMX_NBNXN_SIMD_2XNN``
276 force the use of 2x(N+N) SIMD CPU non-bonded kernels,
277 mutually exclusive of ``GMX_NBNXN_SIMD_4XN``.
279 ``GMX_NBNXN_SIMD_4XN``
280 force the use of 4xN SIMD CPU non-bonded kernels,
281 mutually exclusive of ``GMX_NBNXN_SIMD_2XNN``.
284 disables optimized all-vs-all kernels.
286 ``GMX_NO_CART_REORDER``
287 used in initializing domain decomposition communicators. Rank reordering
288 is default, but can be switched off with this environment variable.
290 ``GMX_NO_LJ_COMB_RULE``
291 force the use of LJ paremeter lookup instead of using combination rules
292 in the non-bonded kernels.
294 ``GMX_NO_CUDA_STREAMSYNC``
295 the opposite of ``GMX_CUDA_STREAMSYNC``. Disables the use of the
296 standard cudaStreamSynchronize-based GPU waiting to improve performance when using CUDA driver API
297 ealier than v5.0 with ECC-enabled GPUs.
299 ``GMX_NO_INT``, ``GMX_NO_TERM``, ``GMX_NO_USR1``
300 disable signal handlers for SIGINT,
301 SIGTERM, and SIGUSR1, respectively.
304 do not use separate inter- and intra-node communicators.
307 skip non-bonded calculations; can be used to estimate the possible
308 performance gain from adding a GPU accelerator to the current hardware setup -- assuming that this is
309 fast enough to complete the non-bonded calculations while the CPU does bonded force and PME computation.
310 Freezing the particles will be required to stop the system blowing up.
313 when set, do not add virial contribution to COM pull forces.
316 shell positions are not predicted.
319 turns off solvent optimizations; automatic if ``GMX_NB_GENERIC``
323 the ideal number of charge groups per neighbor searching grid cell is hard-coded
324 to a value of 10. Setting this environment variable to any other integer value overrides this hard-coded
328 set the number of OpenMP or PME threads (overrides the number guessed by
332 use P3M-optimized influence function instead of smooth PME B-spline interpolation.
334 ``GMX_PME_THREAD_DIVISION``
335 PME thread division in the format "x y z" for all three dimensions. The
336 sum of the threads in each dimension must equal the total number of PME threads (set in
340 if the number of domain decomposition cells is set to 1 for both x and y,
341 decompose PME in one dimension.
343 ``GMX_REQUIRE_SHELL_INIT``
344 require that shell positions are initiated.
346 ``GMX_REQUIRE_TABLES``
347 require the use of tabulated Coulombic
348 and van der Waals interactions.
351 the minimum value for soft-core sigma. **Note** that this value is set
352 using the :mdp:`sc-sigma` keyword in the :ref:`mdp` file, but this environment variable can be used
353 to reproduce pre-4.5 behavior with respect to this parameter.
356 should contain multiple masses used for test particle insertion into a cavity.
357 The center of mass of the last atoms is used for insertion into the cavity.
360 use graph for bonded interactions.
362 ``GMX_VERLET_BUFFER_RES``
363 resolution of buffer size in Verlet cutoff scheme. The default value is
364 0.001, but can be overridden with this environment variable.
367 Not strictly a |Gromacs| environment variable, but on large machines
368 the hwloc detection can take a few seconds if you have lots of MPI processes.
369 If you run the hwloc command `lstopo out.xml` and set this environment
370 variable to point to the location of this file, the hwloc library will use
371 the cached information instead, which can be faster.
374 the ``mpirun`` command used by :ref:`gmx tune_pme`.
377 the :ref:`gmx mdrun` command used by :ref:`gmx tune_pme`.
379 ``GMX_DISABLE_DYNAMICPRUNING``
380 disables dynamic pair-list pruning. Note that :ref:`gmx mdrun` will
381 still tune nstlist to the optimal value picked assuming dynamic pruning. Thus
382 for good performance the -nstlist option should be used.
384 ``GMX_NSTLIST_DYNAMICPRUNING``
385 overrides the dynamic pair-list pruning interval chosen heuristically
386 by mdrun. Values should be between the pruning frequency value
387 (1 for CPU and 2 for GPU) and :mdp:`nstlist` ``- 1``.
389 ``GMX_USE_TREEREDUCE``
390 use tree reduction for nbnxn force reduction. Potentially faster for large number of
391 OpenMP threads (if memory locality is important).
393 .. _opencl-management:
397 Currently, several environment variables exist that help customize some aspects
398 of the OpenCL_ version of |Gromacs|. They are mostly related to the runtime
399 compilation of OpenCL kernels, but they are also used in device selection.
401 ``GMX_OCL_NOGENCACHE``
402 If set, disable caching for OpenCL kernel builds. Caching is
403 normally useful so that future runs can re-use the compiled
404 kernels from previous runs. Currently, caching is always
405 disabled, until we solve concurrency issues.
408 Enable OpenCL binary caching. Only intended to be used for
409 development and (expert) testing as neither concurrency
410 nor cache invalidation is implemented safely!
412 ``GMX_OCL_NOFASTGEN``
413 If set, generate and compile all algorithm flavors, otherwise
414 only the flavor required for the simulation is generated and
417 ``GMX_OCL_DISABLE_FASTMATH``
418 Prevents the use of ``-cl-fast-relaxed-math`` compiler option.
421 If defined, the OpenCL build log is always written to the
422 mdrun log file. Otherwise, the build log is written to the
423 log file only when an error occurs.
426 If defined, it enables verbose mode for OpenCL kernel build.
427 Currently available only for NVIDIA GPUs. See ``GMX_OCL_DUMP_LOG``
428 for details about how to obtain the OpenCL build log.
430 ``GMX_OCL_DUMP_INTERM_FILES``
432 If defined, intermediate language code corresponding to the
433 OpenCL build process is saved to file. Caching has to be
434 turned off in order for this option to take effect (see
435 ``GMX_OCL_NOGENCACHE``).
437 - NVIDIA GPUs: PTX code is saved in the current directory
438 with the name ``device_name.ptx``
439 - AMD GPUs: ``.IL/.ISA`` files will be created for each OpenCL
440 kernel built. For details about where these files are
441 created check AMD documentation for ``-save-temps`` compiler
445 Use in conjunction with ``OCL_FORCE_CPU`` or with an AMD device.
446 It adds the debug flag to the compiler options (-g).
449 Disable optimisations. Adds the option ``cl-opt-disable`` to the
452 ``GMX_OCL_FORCE_CPU``
453 Force the selection of a CPU device instead of a GPU. This
454 exists only for debugging purposes. Do not expect |Gromacs| to
455 function properly with this option on, it is solely for the
456 simplicity of stepping in a kernel and see what is happening.
458 ``GMX_OCL_DISABLE_I_PREFETCH``
459 Disables i-atom data (type or LJ parameter) prefetch allowig
462 ``GMX_OCL_ENABLE_I_PREFETCH``
463 Enables i-atom data (type or LJ parameter) prefetch allowig
464 testing on platforms where this behavior is not default.
466 ``GMX_OCL_NB_ANA_EWALD``
467 Forces the use of analytical Ewald kernels. Equivalent of
468 CUDA environment variable ``GMX_CUDA_NB_ANA_EWALD``
470 ``GMX_OCL_NB_TAB_EWALD``
471 Forces the use of tabulated Ewald kernel. Equivalent
472 of CUDA environment variable ``GMX_OCL_NB_TAB_EWALD``
474 ``GMX_OCL_NB_EWALD_TWINCUT``
475 Forces the use of twin-range cutoff kernel. Equivalent of
476 CUDA environment variable ``GMX_CUDA_NB_EWALD_TWINCUT``
478 ``GMX_OCL_FILE_PATH``
479 Use this parameter to force |Gromacs| to load the OpenCL
480 kernels from a custom location. Use it only if you want to
481 override |Gromacs| default behavior, or if you want to test
484 ``GMX_OCL_DISABLE_COMPATIBILITY_CHECK``
485 Disables the hardware compatibility check. Useful for developers
486 and allows testing the OpenCL kernels on non-supported platforms
487 (like Intel iGPUs) without source code modification.
489 Analysis and Core Functions
490 ---------------------------
492 accuracy in Gaussian L510 (MC-SCF) component program.
494 ``GMX_QM_ORCA_BASENAME``
495 prefix of :ref:`tpr` files, used in Orca calculations
496 for input and output file names.
499 when set to a nonzero value, Gaussian QM calculations will
500 iteratively solve the CP-MCSCF equations.
502 ``GMX_QM_MODIFIED_LINKS_DIR``
503 location of modified links in Gaussian.
506 used by :ref:`gmx do_dssp` to point to the ``dssp``
507 executable (not just its path).
510 directory where Gaussian is installed.
513 name of the Gaussian executable.
515 ``GMX_DIPOLE_SPACING``
516 spacing used by :ref:`gmx dipoles`.
519 sets the maximum number of residues to be renumbered by
520 :ref:`gmx grompp`. A value of -1 indicates all residues should be renumbered.
522 ``GMX_FFRTP_TER_RENAME``
523 Some force fields (like AMBER) use specific names for N- and C-
524 terminal residues (NXXX and CXXX) as :ref:`rtp` entries that are normally renamed. Setting
525 this environment variable disables this renaming.
528 ``gunzip`` executable, used by :ref:`gmx wham`.
531 name of X11 font used by :ref:`gmx view`.
534 the time unit used in output files, can be
535 anything in fs, ps, ns, us, ms, s, m or h.
537 ``GMX_QM_GAUSSIAN_MEMORY``
538 memory used for Gaussian QM calculation.
541 name of the ``multiprot`` executable, used by the
542 contributed program ``do_multiprot``.
545 number of CPUs to be used for Gaussian QM calculation
548 directory where Orca is installed.
551 simulated annealing step size for Gaussian QM calculation.
553 ``GMX_QM_GROUND_STATE``
554 defines state for Gaussian surface hopping calculation.
557 name of the ``total`` executable used by the contributed
558 ``do_shift`` program.
561 make :ref:`gmx energy` and :ref:`gmx eneconv`
565 where to find VMD plug-ins. Needed to be
566 able to read file formats recognized only by a VMD plug-in.
569 base path of VMD installation.
572 sets viewer to ``xmgr`` (deprecated) instead of ``xmgrace``.