Amber Bug Fixes and Updates
Note: The individual bug fixes are described below in case you want
to see what they do, or you don't want to use update_amber for some
reason. Most users are encouraged to use the update_amber script
(which is in $AMBERHOME, and is automatically run during the configure step) to
handle patching tasks. This script automatically downloads and applies patches
on this site individually and keeps track of every patch that has been applied,
making patching and trouble-shooting easier.
Instructions for using
update_amber can be found in the
Amber 2018 Reference Manual in Chapter 2, Section 3. See
Jason's wiki page
for more detailed instructions. Note that
built-in Proxy support (see the Amber Reference Manual for instructions on
setting up your Proxy).
If you wish to apply these updates, run the command
Alternatively, the Amber configure script automatically checks for and
downloads updates so you can just recompile Amber.
Amber18 Updates (April, 2018)
Amber16 Updates (April, 2016)
Adds GPU support for including external electric fields.
Completes GPU support for including external electric fields.
Adds support for Intel Xeon and Intel Xeon Phi Optimizations.
New free energy routines in
nfe.pdf for full documentation.
Support Titan-XP in AMBER 16 pmemd.cuda.
Provides a 10 to 15 percent performance boost to GP102,104 and 107 based
gpus, which include GTX-1060,1070,1080, Titan-XP, Quadro P5000 and P6000.
Also enables execution on P100 based GPUs.
Updates printouts in the code (re SM60) to match those on the gpu page
of the web page.
Fix NetCDF restart writing for
pmemd when internal
coordinate axes are flipped.
Fixes and extensions to the "nfe" options in
Fix vulnerability that allowed Lennard-Jones 12-6-4 potentials to be
applied with incompatible features in
Fixes overwriting output in gamd.log when continuing a simulation.
Fixes dt and nstep passing in gpu force calculations (primarily used in
oscillating electric fields).
Amber14 Updates (April, 2014)
Fixes a bug in
pmemd.amoeba.MPI when running on
multiple processors as well as a small fix in the
pmemd.amoeba softcore code.
Fixes a bug in
pmemd.MPI when running with constant
ntp = 1,
barostat = 1 (default) and a
large number of processors. This bug causes a deviation in the pressure
calculation. Presence of this bug can be identified when running under
constant pressure and
block_fft = 1 appears in the output
Fixes printout of
NMRopt trace variables (distances, angles,
torsions, etc.) when using the Monte Carlo barostat in
pmemd. Also prevents illegal values of mcbarint with
multiple time-stepping integrators.
Fixes incorrect printing of Ewald Error Estimate for CPU based
Adds MIC offload support to
pmemd. This update enables
use of Intel Xeon Phi co-processor cards in offload mode where the card acts
as an accelerator for the direct space sum.
Updates GPU code to v14.0.1 - add support for CUDA 6.5, fixes issues on
> 4 gpus, enables peer to peer support on 8 gpus for systems that support
it in hardware and adds checks for NANs in restrt files.
Fixes a bug in that would cause the number of shake constraints removed to be
reported incorrectly with softcore TI. Add an additional warning if shake is
attempted for bonds that have large differences in equilibirum bond lengths.
Update code to use 80 character line width to fix compile issues with certain
Enhancements have been made to the MIC offload implementation for improved
performance. Please refer to the
documentation for more details.
Center of mass distance restraints now supported in the GPU engine. Usage
follows the same setup and protocol as in the CPU implemetation. For the GPU
implementation all calculations including COM are done on the GPU.
Updates and bug fixes in pmemd.cuda relating to emap, COM restraints and the
GB solvent model:
- emap: A check has been added to prevent restraints when using emap as
they are not yet supported on the GPU.
- COM restraints: The number of atoms in a COM grouping has been
increased from 200 to 1024. A bug has been fixed that was preventing
distance restraints from correctly being detected when the restrain
involved atom 1.
- GB: Fixes undocumented uninitialized variable in DPFP GB simulations
that was causing inconsistent results.
Updates test cases for the new GBneck2 nucleic acid parameters introduced
with AmberTools 15. This should make sure that all Amber14 tests pass after
Fixes GPU-COM distance restraint bug in pmemd.cuda. Adds error checking code
ir6 = 1 when used with GPU-COM distance restraints.
Improves the GPU-COM distance restraint test cases.
Amber12 Updates (March, 2012)
Fix issue where restart files from pmemd can be truncated upon abnormal
This updates the version number of all Amber programs
addles) to report Amber 12.
This turns EMAP reading errors into fatal errors and fixes restart file
printing for pmemd minimizations (completing bugfix.1)
Prevents potential user errors with the QM/MM interface to external QM
packages if the QM region has a charge. Also included are missing
test files for mechanical embedding.
Fix to prevent compilation and linking errors on
- bugfix.6: Sometimes the
temperature was not set from the ASCII restart file when restarting a T-REMD
- bugfix.7: Fix bugfix.1
so you don't get errors opening the restart file if you don't use "-O" on the
- bugfix.8: Remove an
optimization directive that causes sporadic softcore-TI run SHAKE errors
with some compilers.
Large CUDA upgrade patch. This is a major update to the GPU support
in AMBER 12 and also contains minor bug fixes for the CPU version of pmemd.
- Support for Kepler K10 and GTX6XX GPUs
- New, high-performance, lower-memory footprint hybrid fixed precision
- Support for REMD using 1 GPU per replica
- Various bugfixes and tweaks for aMD, IPS, GB and NMR restrained
- bugfix.10: Multiple
bugfixes for regular and softcore TI.
- bugfix.11: Update to
bugfix.10, for the net charge removal code.
- bugfix.12: Fix race
condition in pmemd.cuda.MPI that may cause failures in parallel GPU
- bugfix.13: Fix the
instability problem of H-REMD trajectories (both sander and pmemd) by
swapping velocities as well as coordinates. updating
- bugfix.14: Multiple
fixes and updates for GPU code:
- Adds support for CUDA Toolkit 5.0
- Fixes bugs in the Jarzynski code on GPUs and re-enables it.
- Fixes a bug with the use of harmonic restraints when running with
- Enables GBSA simulations in GPU runs.
- Updates requested citations.
- Fixes uninitialized variables in
NMRopt = 1 runs that
could cause crashes with some compilers.
- Adds check for forgetting to specify gamma_ln with ntb>1 and ntt=3
- Updates GPU code to v.12.2
- bugfix.15: Fixes for
- Add self-plasma term for charged systems
- Improve default values for k space vectors
- Corresponding updates to tests
- bugfix.19: Minor
- Adds support for CUDA 5.5. - via accompanying AMBERTools 13 bugfix.16.
NMROPT on multiple GPUs due to unfixed bugs in
the MPI version at present.
- Enables support for GTX-Titan and GTX780 GPUs and checks that the
minimum required driver version of v325.15 is installed.
- Fix spurious crash with kNLBuildNeighborListOrthogonal16_kernel
for systems containing large vacuum bubbles or very low density.
- Updates GPU code citation.
- Adds missing test case output for GPU Jarinski Tests.
- Updates GPU code to v12.3.1
- bugfix.20: Allow
pmemd.cuda to run with CUDA driver version 319.60 or
- bugfix.21: EXTERN
QM/MM enhancements and bug fixes:
- For Orca:
- Fix bug with forces on MM atoms (update read format for Orca .pcgrad
- Fix problems with parallel Orca runs
- Enable single point trajectory post-processing
- Update dipole moment output to Debye
- For Gaussian:
- QM/MM with MP2 and electrostatic embedding now uses MP2 density to
compute correct MP2 forces on MM atoms
- Allow 256 character route section (avoids truncation)
- Fix problems with Gaussian 03
- For ADF:
- Update dipole moment ouput to Debye
- For all:
- Template input files support 256 characters per line (avoids
- Input of charge and spin multiplicity in the namelist of the external
- QM program has been deprecated - this needs to be given in the
&qmmm namelist (variables qmcharge and spin)
- Includes updates to tests
- Older versions of Amber are no longer supported for bug fixes. If you
are using one of these versions, all of the functionality that you might have
gotten through the
sander program or auxiliary software
are now distributed freely with much never releases of