Re: [AMBER] AMD GPU install problems (David A Case)

From: Derek M Shore via AMBER <>
Date: Thu, 8 Jun 2023 18:39:05 +0000

Hi Dr. Case,

Thanks very much—using AmberTools22 for the install was exactly what was needed! The AMD/HIP-supported compile was trivial with this version of AmberTools.

AmberTool22 is perfectly fine for our purposes—we were only trying AmberTools23 because that’s the only available version of AmberTools currently available on

Thanks again,

Derek M. Shore, Ph.D.
Lecturer of Physiology and Biophysics
Scientific Computing Technology Engineer
Institute for Computational Biomedicine

Weill Cornell Medicine
Department of Physiology and Biophysics
1300 York Avenue, Room LC501A
New York, NY 10065<>

From: <>
Date: Wednesday, June 7, 2023 at 3:00 PM
To: <>
Subject: [EXTERNAL] AMBER Digest, Vol 4074, Issue 1
Send AMBER mailing list submissions to

To subscribe or unsubscribe via the World Wide Web, visit
or, via email, send a message with subject or body 'help' to

You can reach the person managing the list at

When replying, please edit your Subject line so it is more specific
than "Re: Contents of AMBER digest..."

AMBER Mailing List Digest

Today's Topics:

   1. Re: AMD GPU install problems (David A Case)
   2. Re: Problems with minimization energy NaN (David A Case)
   3. Re: Problems with minimization energy NaN (Bhattacharjee, Sinjini)
   4. Trouble Installing OpenMPI on Amazon Linux 2 (CentOS)
      (Evan Biedermann)
   5. H-bond analysis in cpptraj (Prithviraj Nandigrami)
   6. Native contacts analysis in CPPTRAJ (Prithviraj Nandigrami)


Message: 1
Date: Tue, 6 Jun 2023 22:09:17 -0400
From: David A Case <>
To: Charo del Genio <>, AMBER Mailing List
Subject: Re: [AMBER] AMD GPU install problems
Message-ID: <20230607020917.vqxsegtvyteqtaml.pop-os.localdomain>
Content-Type: text/plain; charset=utf-8; format=flowed

On 05/Jun/2023 17:22, Derek M Shore via AMBER wrote:

>I am trying to install Amber22 on the Frontier supercomputer at Oak
>Ridge National Lab. The compute nodes are composed of AMD Optimized 3rd
>Gen EPYC CPUs and AMD MI250X GPUs. I can install AmberTools 23/Amber22
>without GPU support without any problems (and the install passes the test
>suite). However, I run into problems when trying to install with GPU
>I?m following the AMD recipe found here:

We are aware of problems related to AMD and AmberTools23, and are working on
a fix. For the moment, however, for GPUs I think you will need to stick
with the AmberTools22/Amber22 combination. There are very few (if any?) GPU
changes between AmberTools22 and AmberTools23, so you won't really be
missing anything -- you already have a CPU version with AmberTools23, so just
install a GPU version of AmberTools22/Amber22 in a separate PATH.

You can get Amber22 from the Amber web site, and AmberTools22 is available

Let us know if you have problems with this route -- we don't have a lot of
Frontier users (or maybe they are all satisfied?). Others might chime in

....hope this helps....dac


Message: 2
Date: Wed, 7 Jun 2023 08:27:59 -0400
From: David A Case <>
To: "Bhattacharjee, Sinjini" <>
Subject: Re: [AMBER] Problems with minimization energy NaN
Message-ID: <20230607122759.c3xvcspxsdclsqpj.pop-os.localdomain>
Content-Type: text/plain; charset=us-ascii; format=flowed

On Wed, Jun 07, 2023, Bhattacharjee, Sinjini wrote:
>1. As a first step I tried only setting up the protein and non-standard
>residues (eg. Hemes, FES clusters), without the lipid bilayer.
>2. I used check in cpptraj to remove all bad contacts and based on the atom
>name printed in NSTEP1 manually.
>3. Then I do minimization with cpu version of sander in Amber20 and Still
>it gives me the following error:
> 1 NaN NaN 4.1424E+03 HB2 6514
> BOND = 7826.5578 ANGLE = 6795.2519 DIHED = 24953.4886
> VDWAALS = NaN EEL = -1339783.3680 HBOND = 0.0000
> 1-4 VDW = NaN 1-4 EEL = 80816.4521 RESTRAINT = 0.0000

You are describing what your intent was, not what you actually did.

1. How (exactly) did you manually remove bad contacts? How many were
removed? (As I suggested before, start just with those with zero non-bonded
distance.) Did they involve hydrogens?

2. Did you then (a) re-run tleap to build a new system; and (b) re-run
cpptraj check to re-check for bad contacts?

A guess: if you just removed certain hydrogen atoms (say), they will get
re-built by tleap (in the same manner as they were in your initial run.)
That doesn't get you anywhere. You will need to genuinely move some atoms
to better locations.

>Now, when I check the structure, atom 6514 (HB2) is not overlapping with
>anything! The closest non-bonded atom is at 1.2 Ang (fig attached). Then
>what is the source of this error and how should I avoid this further? Is it
>due to the charges?

The GMAX value just prints the largest element of the force vector. But if
some of the elements are NaN (as in your case), this information is not of
much use, since NaN's are ignored in the search for the maximum value
(but not in the RMS calculation.)

...good luck...dac


Message: 3
Date: Wed, 7 Jun 2023 12:49:51 +0000
From: "Bhattacharjee, Sinjini" <>
To: "" <>
Subject: Re: [AMBER] Problems with minimization energy NaN
Message-ID: <>
Content-Type: text/plain; charset="utf-8"

Hi all,

I think I found the culplrit. It was not overlaps but a mess up in the vdw term of a non-standard residue, the parameters of which I imported from CHARMM. All good now, the minimization works :)

Thanks for all your help and suggestions!
Sinjini Bhattacharjee

PhD Student
Molecular Theory and Spectroscopy
Max-Planck-Institut f?r Kohlenforschung
45470 M?lheim an der Ruhr, Germany

On 2. Jun 2023, at 16:26, Bhattacharjee, Sinjini <<>> wrote:
Hi, I am trying to minimise a protein-membrane system using sander and pmemd in Amber20. However, the calculation keep crashing in the first cycle with the following error:
NSTEP       ENERGY          RMS            GMAX         NAME    NUMBER
      1              NaN            NaN     1.0669E+24     C19     90066
 BOND    =    31692.7352  ANGLE   =    68711.0955  DIHED      =    59597.8979
 VDWAALS =           NaN  EEL     =   -12389.2862  HBOND      =        0.0000
 1-4 VDW =           NaN  1-4 EEL =  -149672.0786  RESTRAINT  =        0.0000
This is my initial input:
 System minimization:
   imin=1, ntmin=1, nmropt=0
   maxcyc=2000, ncyc=1000,
   ntx=1, irest=0,
   ntpr=100, ntwr=100, iwrap=0,
   ntf=1, ntb=1, cut=10.0, nsnb=20,
   ibelly=0, ntr=1,
   restraintmask=?.CA", restraint_wt=20.0,
I added the membrane using packmol-memgen.
The system size is huge and I checked and removed clashes in the protein and the non-standard residues.
Can you please tell me what is the problem and how can I solve this? Is it an issue with the parameters or the structure?
Many thanks,
Sinjini Bhattacharjee
PhD Student
Molecular Theory and Spectroscopy
Max-Planck-Institut f?r Kohlenforschung
45470 M?lheim an der Ruhr, Germany
Message: 4
Date: Wed, 7 Jun 2023 13:12:40 -0400
From: Evan Biedermann <>
Subject: [AMBER] Trouble Installing OpenMPI on Amazon Linux 2 (CentOS)
Content-Type: text/plain; charset="UTF-8"
*Problem Summary:*
I am having an issue installing Amber22 with CUDA and MPI. I have followed
the directions for installing in centos using cmake following both  and .
I ran the initial install following the step 3 of page 23 on the pdf
instructions. Then I installed the MPI compilers in the
$AMBERHOME/AmberTools/src path as specified in step 8 of page 24. After
that, I went back to the run_cmake file, set DMPI=TRUE, and ran sudo
The mpifort and mpif90 compilers are in the same path as the MPI C and MPI
CXX Compilers, but for some reason cmake is unable to find them. I have
tried adding the path to .bashrc, adding set(MPI_Fortran_COMPILER
/opt/amber22/AmberTools/src/bin), set(MPI_Fortran_INCLUDE_PATH
/opt/amber22/AmberTools/src/bin), set(MPI_Fortran_LIBRARIES
/opt/amber22/AmberTools/src/bin) to the MPIConfig.cmake and CMakeList.txt
files and nothing has worked. I have also tried setting the INCLUDE_PATH
and LIBRARIES to  /opt/amber22/AmberTools/src/lib/openmpi.
To add more specifics, I originally installed the version of OpenMPI by
using wget to download openmpi-4.1.5.tar.bz2 to $AMBERHOME/AmberTools/src,
unpacking and installing following these commands:
   1. *tar -xzf openmpi-4.0.1.tar.gz*
   2. *cd openmpi-4.1.5*
*./configure --prefix=/opt/amber22/AmberTools/src *
   4. *make*
   5. *sudo make all install*
This ended up with the following bin folder in this
path: /opt/amber22/AmberTools/src/bin
[image: image.png]
(if the picture does load, the bin contains mpicc, mpicxx, mpif77, mpirun,
mpiCC, mpiexec, mpifort, mpif90, mpic++....)
Here are the full details on the system I'm using:
*NAME="Amazon Linux"VERSION="2"ID="amzn"ID_LIKE="centos rhel
fedora"VERSION_ID="2"PRETTY_NAME="Amazon Linux
< >"*
*Description:    Amazon Linux release 2 (Karoo)*
I do believe that the issue is coming from a base version of mpi
interfering with the one I tried to install, because originally when I used
the command *which mpif90* it would show a path to */efs/anaconda3/bin*. I
then tried *sudo yum install openmpi-devel* and *module load mpi* and
now *which
mpif90* gives */opt/amazon/openmpi/bin/mpif90* and *mpif90 --version*
*GNU Fortran (GCC) 7.3.1 20180712 (Red Hat 7.3.1-15)*
*Copyright (C) 2017 Free Software Foundation, Inc.This is free software;
see the source for copying conditions.  There is NOwarranty; not even for
However, the Amber build is still failing giving the same output as the
original error:
Starting configuration of Amber version 22.0.0...-- CMake Version: 3.26.4--
For how to use this build system, please read this wiki:--
< >-- For a list of important
CMake variables, check here:--
< >--
Amber source found, building AmberTools and Amber-- Could NOT find
MPI_Fortran (missing: MPI_Fortran_WORKS)-- Could NOT find MPI (missing:
MPI_Fortran_FOUND) (found version "4.0")-- MPI C Compiler:
/opt/amber22/AmberTools/src/bin/-- MPI CXX Compiler:
/opt/amber22/AmberTools/src/bin/CMake Error at cmake/MPIConfig.cmake:16
(message):  You requested MPI, but the MPI Fortran library was not found.
Please  install one and try again, or set MPI_Fortran_INCLUDE_PATH and
MPI_Fortran_LIBRARIES to point to your MPI.Call Stack (most recent call
first):  CMakeLists.txt:120 (include)-- Configuring incomplete, errors
occurred!If errors are reported, search for 'CMake Error' in the cmake.log
file.If the cmake build report looks OK, you should now do the following:
  make install    source /opt/amber22/amber.shConsider adding the last line
to your login startup script, e.g. ~/.bashrc*
It looks like the MPI C and MPI CC compilers are still being sourced from
/opt/amber22/AmberTools/src/bin/ but it can't find the fortran compiler.
I've tried removing /opt/amber22/AmberTools/src/bin/ from the PATH to use
the yum install, but the build is still
sourcing /opt/amber22/AmberTools/src/bin/ for MPI C and MPI CC.
Please let me know how I can resolve this issue.
Evan Biedermann
Associate Solutions Architect
Engineering | Cloud303
12081 W Alameda Pkwy, Suite 150, Lakewood, CO 80228
< >
Message: 5
Date: Wed, 7 Jun 2023 14:46:53 -0400
From: Prithviraj Nandigrami <>
To: AMBER Mailing List <>
Subject: [AMBER] H-bond analysis in cpptraj
Content-Type: text/plain; charset="UTF-8"
Dear AMBER Users and Experts,
Could someone please guide me on how to print out the total number of
hydrogen bonds formed between solute-solute (excluding the solvent) in a MD
I am using the following script:
parm input.parm7
hbond Backbone :1-90.C,O,N,H avgout BB.avg.dat series uuseries bbhbond.gnu
How do I calculate the total number of H-bonds formed in each frame
during the course of an MD run?
Many thanks for your help in advance.
Message: 6
Date: Wed, 7 Jun 2023 14:56:41 -0400
From: Prithviraj Nandigrami <>
To: AMBER Mailing List <>
Subject: [AMBER] Native contacts analysis in CPPTRAJ
Content-Type: text/plain; charset="UTF-8"
Dear AMBER Users and Experts,
I am trying to use the native contacts analysis script implemented in
CPPTRAJ for a couple of MD trajectories as follows:
parm input.strip.parm7
trajin < >
nativecontacts name NC1 :1-85&!.H= \
   writecontacts native-contacts.dat \
   resout resout.dat \
   distance 5.0 \
   byresidue out all-residues.dat mindist maxdist \
   map mapout gnu \
   contactpdb contactspdb.pdb \
   series seriesout native-contacts-series.dat
lifetime NC1[NC] out lifetime.dat
 Here is an example for native.gnu file:
   1.000    1.000       0.0000
   1.000    2.000      16.6333
   1.000    3.000       0.0000
   1.000    4.000       0.0000
   1.000    5.000       0.0000
   1.000    6.000       0.0000
   1.000    7.000       0.0000
   1.000    8.000       0.0000
   1.000    9.000       0.0000
   1.000   10.000       0.0000
I have two sets of MD runs - wild-type protein and mutant protein. I used
the script above to calculate native.gnu files for these two systems.
I was looking to plot a "difference" plot for these two systems with the
hope that regions with the most significant differences would show up.
What is the correct way to go about this? If I take the difference of the
third column (normalized) for the wild-type and mutant cases, and try to
plot a contact map, is that it?
I would appreciate any help or guidance.
Thank you.
Subject: Digest Footer
AMBER mailing list
End of AMBER Digest, Vol 4074, Issue 1
AMBER mailing list
Received on Thu Jun 08 2023 - 12:00:02 PDT
Custom Search