Amber11 Bug Fixes
11.0 April, 2010
The bug fixes detailed below are contained in a
single patch tarball. You can patch your Amber11 distribution by doing the
NOTE: If you have already applied the major CUDA GPU feature enhancement bugfix
(bugfix.17), then you can apply all future bug fixes by hand according to the
instructions in the descriptions of those bug fixes. That will avoid having to
download the first 17 patches again and having to decompress and untar the files.
- Download bugfix.all.tar.bz2.
- Download apply_bugfix.x.
- Be sure that your AMBERHOME environment variable points to the
amber11 directory that contains your Amber11 distribution
- Patch your system by typing the following (in whatever directory you
downloaded these files to):
chmod 700 apply_bugfix.x
This should patch your source files with the latest bugfixes. Any
bugfixes that appear to have already been applied will be skipped.
- Now you can go to the $AMBERHOME/AmberTools/src directory, type
./configure --help, and continue with the installation. Be sure to
follow instructions about running the AT15_Amber11.py script before
- EDIT: Since bugfix.24 for AmberTools 1.5, the above step is no longer
necessary. AT15_Amber11.py is run as part of configure (at the very end)
unless this is explicitly disabled with the -nopatch flag. Without this patch,
and if you use -nopatch, the above step is still required!
Note: The individual bug fixes are described below in case you want to
see what they do, or need to apply just some of them. Most users should use
the apply_bugfix.x script described above.
Fix possible imaging issues for QMMM calculations using very small boxes.
Fix problems when water molecules were included as part of a softcore region. Be sure to apply bugfix.15 as well.
Do not allow es_cutoff /= vdw_cutoff in pmemd.CUDA.
Fix possible segfault when running with iwrap=1 in pmemd.CUDA.
Update bugfix.4 to also support ntb=2.
Exit gracefully if user asks for GB or PB with PUPIL: this combination
is not supported.
Fix mm_pbsa: clean finish, greater 255 threads, space tolerance
Certain versions of the Intel Compiler may not optimize PMEMD correctly at link time leading to poor performance.
This is a large scale bug and feature patch for the AMBER 11 NVIDIA GPU (CUDA)
code developed in close collaboration with NVIDIA. Specifically pmemd.cuda. The patch
addresses a number of outstanding bugs, includes some minor single GPU performance
improvements, and most importantly, through funding provided by NSF Grant 1047875 and
UC Lab award 09-LR-06-117792-WALR, provides support for MPI runs over multiple GPUs.
For updated installation instructions and performance numbers please see
The alpha, beta, and gamma OBC parameters were never initialized for hybrid REMD methods.
This affects the cases hybridgb=2, 5, but hybridgb=1 works as expected.
CUDA: Segfaults and/or shake failures could occur when running NPT calculations, with truncated
octahedron boxes and restraints on molecules identified to be part of the solvent. This bug
fix addresses this. It also restricts the selection of FFT grid dimensions for GPU runs to
match those that provide optimum performance with NVCC v3.2.
CUDA: Fixes several recently discovered bugs in the CUDA code. This includes problems when using
cutoffs or .gt. 8.0 when running with small boxes. Fixes problems with large simulations, such
as the cellulose benchmark dying randomly with an allocation failure and also provides a possible
workaround for simulation hangs experiences on GTX4XX and GTX5XX series cards.
TI requires identical trajectories between both groups in multisander.
If the random seed is different for each group, then the coordinates
diverge from one another and the simulation blows up. The same thing
happens if the user inputs different seeds. Fix this by broadcasting
the worldrank==0's ig value to all other commsander masters in mdread.f
A bug in keeping running averages accumulates the wrong total
Epot on one of the nodes. This only occurs when ntave .ne. 0.
The fix also changes the expected output for a softcore test.
This completes bugfix.2 which allows atoms with a negative ic
type to be flagged soft core. The fix now checks for that
possibility and correctly ignores 10-12 flagged vdW interactions.
Previously, this could lead to unallocated values added to the
potential. The problem mainly concerns decoupling water molecules.
Be sure to apply bugfix.2 first.
For some circumstances with "use_sander=1" in MMPBSA.py, a bug of
sander would prevent it from finishing all the trajectory frames.
This is something to do with "imin == 5 " option and PBSA module.
This fix introduces the fix from the developer's tree that Irvine
team already pushed several months ago.
This bugfix is included in part of the tarball
and does not exist as a standalone file. It must be applied as described
at the top of this page. This updates the PMEMD CUDA GPU support from v2.1
to 2.2. This update effectively doubles PME performance across the board and
addresses a number of bugs including issues with running NPT calculations
with restraints and running simulations with organic solvents. It additionally
adds support for extra points in the GPU code. A complete list of items addressed
by this update is available here.
This bugfix is also included in the tarball
bugfix.all.tar.bz2, but it can
be applied directly using patch -p0 -N < bugfix.18 from AMBERHOME.
This patch fixes the calculation of 10-12 non-bonded terms in sander and pmemd.
This patch fixes weird many-millions of steps crashes for large
(9A and up) nonbond cutoffs.
This bugfix is also included in the tarball.
bugfix.all.tar.bz2, but it can
be applied directly using patch -p0 -N < bugfix.19 from AMBERHOME.
- bugfix.20: Fix for multiple GPU issues,
including NaN results when iwrap=1 and a possible hang at MPI_Win_fence() caused by
vacuum bubbles in the solvent. This bugfix is also included in the tarball.
bugfix.all.tar.bz2, but it can
be applied directly using patch -p0 -N < bugfix.20 from AMBERHOME.
- bugfix.21: Fix issues with running NPT calculations
with Chamber prmtops on GPUs.
- bugfix.22: Fixes problems with restart files
being truncated in pmemd / pmemd.cuda on certain linux installations.
- bugfix.23: Completes bugfix.22 so restart file
writing behavior is correct.