1 .. NOTE: Below is a useful bash one-liner to verify whether there are variables in this file
2 .. no longer present in the code.
3 .. ( export INPUT_FILE='docs/user-guide/environment-variables.rst' GIT_PAGER="cat "; for s in $(grep '^`' $INPUT_FILE | sed 's/`//g' | sed 's/,/ /g'); do count=$(git grep $s | grep -v $INPUT_FILE | wc -l); [ $count -eq 0 ] && printf "%-30s%s\n" $s $count; done ; )
4 .. Another useful one-liner to find undocumentedvariables:
5 .. ( export INPUT_FILE=docs/user-guide/environment-variables.rst; GIT_PAGER="cat "; for ss in `for s in $(git grep getenv | sed 's/.*getenv("\(.*\)".*/\1/' | sort -u | grep '^[A-Z]'); do [ $(grep $s $INPUT_FILE -c) -eq 0 ] && echo $s; done `; do git grep $ss ; done )
10 |Gromacs| programs may be influenced by the use of
11 environment variables. First of all, the variables set in
12 the ``GMXRC`` file are essential for running and
13 compiling |Gromacs|. Some other useful environment variables are
14 listed in the following sections. Most environment variables function
15 by being set in your shell to any non-NULL value. Specific
16 requirements are described below if other values need to be set. You
17 should consult the documentation for your shell for instructions on
18 how to set environment variables in the current shell, or in configuration
19 files for future shells. Note that requirements for exporting
20 environment variables to jobs run under batch control systems vary and
21 you should consult your local documentation for details.
26 |Gromacs| automatically backs up old
27 copies of files when trying to write a new file of the same
28 name, and this variable controls the maximum number of
29 backups that will be made, default 99. If set to 0 it fails to
30 run if any output file already exists. And if set to -1 it
31 overwrites any output file without making a backup.
34 if this is explicitly set, no cool quotes
35 will be printed at the end of a program.
38 prevent dumping of step files during
39 (for example) blowing up during failure of constraint
43 dump all configurations to a :ref:`pdb`
44 file that have an interaction energy less than the value set
45 in this environment variable.
48 ``GMX_VIEW_EPS`` and ``GMX_VIEW_PDB``, commands used to
49 automatically view :ref:`xvg`, :ref:`eps`
50 and :ref:`pdb` file types, respectively; they default to ``xmgrace``,
51 ``ghostview`` and ``rasmol``. Set to empty to disable
52 automatic viewing of a particular file type. The command will
53 be forked off and run in the background at the same priority
54 as the |Gromacs| tool (which might not be what you want).
55 Be careful not to use a command which blocks the terminal
56 (e.g. ``vi``), since multiple instances might be run.
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 and SYCL. Note that CUDA
83 timings 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.
93 number of steps that elapse between dumping
94 the current DD to a PDB file (default 0). This only takes effect
95 during domain decomposition, so it should typically be
96 0 (never), 1 (every DD phase) or a multiple of :mdp:`nstlist`.
98 ``GMX_DD_NST_DUMP_GRID``
99 number of steps that elapse between dumping
100 the current DD grid to a PDB file (default 0). This only takes effect
101 during domain decomposition, so it should typically be
102 0 (never), 1 (every DD phase) or a multiple of :mdp:`nstlist`.
105 general debugging trigger for every domain
106 decomposition (default 0, meaning off). Currently only checks
107 global-local atom index mapping for consistency.
110 over-ride the number of DD pulses used
111 (default 0, meaning no over-ride). Normally 1 or 2.
113 ``GMX_DISABLE_ALTERNATING_GPU_WAIT``
114 disables the specialized polling wait path used to wait for the PME and nonbonded
115 GPU tasks completion to overlap to do the reduction of the resulting forces that
116 arrive first. Setting this variable switches to the generic path with fixed waiting
119 ``GMX_TEST_REQUIRED_NUMBER_OF_DEVICES``
120 sets the number of GPUs required by the test suite. By default, the test suite would
121 fall-back to using CPU if GPUs could not be detected. Set it to a positive integer value
122 to ensure that at least this at least this number of usable GPUs are detected. Default:
123 0 (not testing GPU availability).
125 There are a number of extra environment variables like these
126 that are used in debugging - check the code!
128 Performance and Run Control
129 ---------------------------
130 ``GMX_DO_GALACTIC_DYNAMICS``
131 planetary simulations are made possible (just for fun) by setting
132 this environment variable, which allows setting :mdp:`epsilon-r` to -1 in the :ref:`mdp`
133 file. Normally, :mdp:`epsilon-r` must be greater than zero to prevent a fatal error.
134 See webpage_ for example input files for a planetary simulation.
136 ``GMX_BONDED_NTHREAD_UNIFORM``
137 Value of the number of threads per rank from which to switch from uniform
138 to localized bonded interaction distribution; optimal value dependent on
139 system and hardware, default value is 4.
141 ``GMX_GPU_NB_EWALD_TWINCUT``
142 force the use of twin-range cutoff kernel even if :mdp:`rvdw` equals
143 :mdp:`rcoulomb` after PP-PME load balancing. The switch to twin-range kernels is automated,
144 so this variable should be used only for benchmarking.
146 ``GMX_GPU_NB_ANA_EWALD``
147 force the use of analytical Ewald kernels. Should be used only for benchmarking.
149 ``GMX_GPU_NB_TAB_EWALD``
150 force the use of tabulated Ewald kernels. Should be used only for benchmarking.
152 ``GMX_DISABLE_CUDA_TIMING``
153 Deprecated. Use ``GMX_DISABLE_GPU_TIMING`` instead.
156 perform domain decomposition halo exchange communication operations (on coordinate and force buffers)
157 directly on GPU memory spaces, without the staging of data through CPU memory, where possible.
159 ``GMX_GPU_PME_PP_COMMS``
160 when the simulation uses a separate PME rank, perform communication operations between PP and PME rank
161 (for coordinate and force buffers) directly on GPU memory spaces, without the staging of data through CPU
162 memory, where possible.
164 ``GMX_GPU_SYCL_NO_SYNCHRONIZE``
165 disable synchronizations between different GPU streams in SYCL build, instead relying on SYCL runtime to
166 do scheduling based on data dependencies. Experimental.
169 times all code during runs. Incompatible with threads.
171 ``GMX_CYCLE_BARRIER``
172 calls MPI_Barrier before each cycle start/stop call.
175 build domain decomposition cells in the order
176 (z, y, x) rather than the default (x, y, z).
178 ``GMX_DD_USE_SENDRECV2``
179 during constraint and vsite communication, use a pair
180 of ``MPI_Sendrecv`` calls instead of two simultaneous non-blocking calls
181 (default 0, meaning off). Might be faster on some MPI implementations.
183 ``GMX_DLB_BASED_ON_FLOPS``
184 do domain-decomposition dynamic load balancing based on flop count rather than
185 measured time elapsed (default 0, meaning off).
186 This makes the load balancing reproducible, which can be useful for debugging purposes.
187 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.
189 ``GMX_DLB_MAX_BOX_SCALING``
190 maximum percentage box scaling permitted per domain-decomposition
191 load-balancing step (default 10)
193 ``GMX_DD_RECORD_LOAD``
194 record DD load statistics for reporting at end of the run (default 1, meaning on)
196 ``GMX_DETAILED_PERF_STATS``
197 when set, print slightly more detailed performance information
198 to the :ref:`log` file. The resulting output is the way performance summary is reported in versions
199 4.5.x and thus may be useful for anyone using scripts to parse :ref:`log` files or standard output.
201 ``GMX_DISABLE_SIMD_KERNELS``
202 disables architecture-specific SIMD-optimized (SSE2, SSE4.1, AVX, etc.)
203 non-bonded kernels thus forcing the use of plain C kernels.
205 ``GMX_DISABLE_GPU_TIMING``
206 timing of asynchronously executed GPU operations can have a
207 non-negligible overhead with short step times. Disabling timing can improve performance in these cases.
208 Timings are disabled by default with CUDA and SYCL.
210 ``GMX_DISABLE_GPU_DETECTION``
211 when set, disables GPU detection even if :ref:`gmx mdrun` was compiled
214 ``GMX_DISRE_ENSEMBLE_SIZE``
215 the number of systems for distance restraint ensemble
216 averaging. Takes an integer value.
219 emulate GPU runs by using algorithmically equivalent CPU reference code instead of
220 GPU-accelerated functions. As the CPU code is slow, it is intended to be used only for debugging purposes.
223 disable exiting upon encountering a corrupted frame in an :ref:`edr`
224 file, allowing the use of all frames up until the corruption.
227 update forces when invoking ``mdrun -rerun``.
229 ``GMX_FORCE_UPDATE_DEFAULT_GPU``
230 Force update to run on the GPU by default, overriding the ``mdrun -update auto`` option. Works similar to setting
231 ``mdrun -update gpu``, but (1) falls back to the CPU code-path, if set with input that is not supported and
232 (2) can be used to run update on GPUs in multi-rank cases. The latter case should be
233 considered experimental since it lacks substantial testing. Also, GPU update is only supported with the GPU direct
234 communications and ``GMX_FORCE_UPDATE_DEFAULT_GPU`` variable should be set simultaneously with ``GMX_GPU_DD_COMMS``
235 and ``GMX_GPU_PME_PP_COMMS`` environment variables in multi-rank case. Does not override ``mdrun -update cpu``.
238 set in the same way as ``mdrun -gpu_id``, ``GMX_GPU_ID``
239 allows the user to specify different GPU IDs for different ranks, which can be useful for selecting different
240 devices on different compute nodes in a cluster. Cannot be used in conjunction with ``mdrun -gpu_id``.
243 set in the same way as ``mdrun -gputasks``, ``GMX_GPUTASKS`` allows the mapping
244 of GPU tasks to GPU device IDs to be different on different ranks, if e.g. the MPI
245 runtime permits this variable to be different for different ranks. Cannot be used
246 in conjunction with ``mdrun -gputasks``. Has all the same requirements as ``mdrun -gputasks``.
248 ``GMX_GPU_DISABLE_COMPATIBILITY_CHECK``
249 Disables the hardware compatibility check in OpenCL and SYCL. Useful for developers
250 and allows testing the OpenCL/SYCL kernels on non-supported platforms without source code modification.
252 ``GMX_IGNORE_FSYNC_FAILURE_ENV``
253 allow :ref:`gmx mdrun` to continue even if
257 when set to a floating-point value, overrides the default tolerance of
258 1e-5 for force-field floating-point parameters.
260 ``GMX_MAXCONSTRWARN``
261 if set to -1, :ref:`gmx mdrun` will
262 not exit if it produces too many LINCS warnings.
265 neighbor list balancing parameter used when running on GPU. Sets the
266 target minimum number pair-lists in order to improve multi-processor load-balance for better
267 performance with small simulation systems. Must be set to a non-negative integer,
268 the 0 value disables list splitting.
269 The default value is optimized for supported GPUs
270 therefore changing it is not necessary for normal usage, but it can be useful on future architectures.
273 when set, print detailed neighbor search cycle counting.
275 ``GMX_NBNXN_EWALD_ANALYTICAL``
276 force the use of analytical Ewald non-bonded kernels,
277 mutually exclusive of ``GMX_NBNXN_EWALD_TABLE``.
279 ``GMX_NBNXN_EWALD_TABLE``
280 force the use of tabulated Ewald non-bonded kernels,
281 mutually exclusive of ``GMX_NBNXN_EWALD_ANALYTICAL``.
283 ``GMX_NBNXN_SIMD_2XNN``
284 force the use of 2x(N+N) SIMD CPU non-bonded kernels,
285 mutually exclusive of ``GMX_NBNXN_SIMD_4XN``.
287 ``GMX_NBNXN_SIMD_4XN``
288 force the use of 4xN SIMD CPU non-bonded kernels,
289 mutually exclusive of ``GMX_NBNXN_SIMD_2XNN``.
291 ``GMX_NOOPTIMIZEDKERNELS``
292 deprecated, use ``GMX_DISABLE_SIMD_KERNELS`` instead.
294 ``GMX_NO_CART_REORDER``
295 used in initializing domain decomposition communicators. Rank reordering
296 is default, but can be switched off with this environment variable.
298 ``GMX_NO_LJ_COMB_RULE``
299 force the use of LJ paremeter lookup instead of using combination rules
300 in the non-bonded kernels.
302 ``GMX_NO_INT``, ``GMX_NO_TERM``, ``GMX_NO_USR1``
303 disable signal handlers for SIGINT,
304 SIGTERM, and SIGUSR1, respectively.
307 do not use separate inter- and intra-node communicators.
310 skip non-bonded calculations; can be used to estimate the possible
311 performance gain from adding a GPU accelerator to the current hardware setup -- assuming that this is
312 fast enough to complete the non-bonded calculations while the CPU does bonded force and PME computation.
313 Freezing the particles will be required to stop the system blowing up.
315 ``GMX_PULL_PARTICIPATE_ALL``
316 disable the default heuristic for when to use a separate pull MPI communicator (at >=32 ranks).
319 shell positions are not predicted.
321 ``GMX_NO_UPDATEGROUPS``
322 turns off update groups. May allow for a decomposition of more
323 domains for small systems at the cost of communication during update.
325 ``GMX_PME_NUM_THREADS``
326 set the number of OpenMP or PME threads; overrides the default set by
327 :ref:`gmx mdrun`; can be used instead of the ``-npme`` command line option,
328 also useful to set heterogeneous per-process/-node thread count.
331 use P3M-optimized influence function instead of smooth PME B-spline interpolation.
333 ``GMX_PME_THREAD_DIVISION``
334 PME thread division in the format "x y z" for all three dimensions. The
335 sum of the threads in each dimension must equal the total number of PME threads (set in
336 :envvar:`GMX_PME_NTHREADS`).
339 if the number of domain decomposition cells is set to 1 for both x and y,
340 decompose PME in one dimension.
342 ``GMX_REQUIRE_SHELL_INIT``
343 require that shell positions are initiated.
346 should contain multiple masses used for test particle insertion into a cavity.
347 The center of mass of the last atoms is used for insertion into the cavity.
349 ``GMX_VERLET_BUFFER_RES``
350 resolution of buffer size in Verlet cutoff scheme. The default value is
351 0.001, but can be overridden with this environment variable.
354 Not strictly a |Gromacs| environment variable, but on large machines
355 the hwloc detection can take a few seconds if you have lots of MPI processes.
356 If you run the hwloc command :command:`lstopo out.xml` and set this environment
357 variable to point to the location of this file, the hwloc library will use
358 the cached information instead, which can be faster.
361 the ``mpirun`` command used by :ref:`gmx tune_pme`.
364 the :ref:`gmx mdrun` command used by :ref:`gmx tune_pme`.
366 ``GMX_DISABLE_DYNAMICPRUNING``
367 disables dynamic pair-list pruning. Note that :ref:`gmx mdrun` will
368 still tune nstlist to the optimal value picked assuming dynamic pruning. Thus
369 for good performance the -nstlist option should be used.
371 ``GMX_NSTLIST_DYNAMICPRUNING``
372 overrides the dynamic pair-list pruning interval chosen heuristically
373 by mdrun. Values should be between the pruning frequency value
374 (1 for CPU and 2 for GPU) and :mdp:`nstlist` ``- 1``.
376 .. _opencl-management:
380 Currently, several environment variables exist that help customize some aspects
381 of the OpenCL_ version of |Gromacs|. They are mostly related to the runtime
382 compilation of OpenCL kernels, but they are also used in device selection.
385 Enable OpenCL binary caching. Only intended to be used for
386 development and (expert) testing as neither concurrency
387 nor cache invalidation is implemented safely!
389 ``GMX_OCL_NOFASTGEN``
390 If set, generate and compile all algorithm flavors, otherwise
391 only the flavor required for the simulation is generated and
394 ``GMX_OCL_DISABLE_FASTMATH``
395 Prevents the use of ``-cl-fast-relaxed-math`` compiler option.
396 Note: fast math is always disabled on Intel devices due to instability.
399 If defined, the OpenCL build log is always written to the
400 mdrun log file. Otherwise, the build log is written to the
401 log file only when an error occurs.
404 If defined, it enables verbose mode for OpenCL kernel build.
405 Currently available only for NVIDIA GPUs. See ``GMX_OCL_DUMP_LOG``
406 for details about how to obtain the OpenCL build log.
408 ``GMX_OCL_DUMP_INTERM_FILES``
410 If defined, intermediate language code corresponding to the
411 OpenCL build process is saved to file. Caching has to be
412 turned off in order for this option to take effect.
414 - NVIDIA GPUs: PTX code is saved in the current directory
415 with the name ``device_name.ptx``
416 - AMD GPUs: ``.IL/.ISA`` files will be created for each OpenCL
417 kernel built. For details about where these files are
418 created check AMD documentation for ``-save-temps`` compiler
422 Use in conjunction with ``OCL_FORCE_CPU`` or with an AMD device.
423 It adds the debug flag to the compiler options (-g).
426 Disable optimisations. Adds the option ``cl-opt-disable`` to the
429 ``GMX_OCL_FORCE_CPU``
430 Force the selection of a CPU device instead of a GPU. This
431 exists only for debugging purposes. Do not expect |Gromacs| to
432 function properly with this option on, it is solely for the
433 simplicity of stepping in a kernel and see what is happening.
435 ``GMX_OCL_DISABLE_I_PREFETCH``
436 Disables i-atom data (type or LJ parameter) prefetch allowing
439 ``GMX_OCL_ENABLE_I_PREFETCH``
440 Enables i-atom data (type or LJ parameter) prefetch allowing
441 testing on platforms where this behavior is not default.
443 ``GMX_OCL_FILE_PATH``
444 Use this parameter to force |Gromacs| to load the OpenCL
445 kernels from a custom location. Use it only if you want to
446 override |Gromacs| default behavior, or if you want to test
449 ``GMX_OCL_SHOW_DIAGNOSTICS``
450 Use Intel OpenCL extension to show additional runtime performance
453 Analysis and Core Functions
454 ---------------------------
457 used by :ref:`gmx do_dssp` to point to the ``dssp``
458 executable (not just its path).
460 ``GMX_DIPOLE_SPACING``
461 spacing used by :ref:`gmx dipoles`.
464 sets the maximum number of residues to be renumbered by
465 :ref:`gmx grompp`. A value of -1 indicates all residues should be renumbered.
467 ``GMX_NO_FFRTP_TER_RENAME``
468 Some force fields (like AMBER) use specific names for N- and C-
469 terminal residues (NXXX and CXXX) as :ref:`rtp` entries that are normally renamed. Setting
470 this environment variable disables this renaming.
473 name of X11 font used by :ref:`gmx view`.
476 the time unit used in output files, can be
477 anything in fs, ps, ns, us, ms, s, m or h.
481 make :ref:`gmx energy` and :ref:`gmx eneconv`
485 where to find VMD plug-ins. Needed to be
486 able to read file formats recognized only by a VMD plug-in.
489 base path of VMD installation.
492 sets viewer to ``xmgr`` (deprecated) instead of ``xmgrace``.