Amber 18 with PLUMED 2

[Qinghua Liao]

Hello PLUMEDers,

I successfully compiled Amber 18 together with PLUMED 2 (github) on one cluster,

and I performed metadynamics simulations using pmemd.cuda (GPU, CUDA-10.1).

I then tried to install them on another cluster using the same steps (CUDA-9.0),

the installation was successful, but I could not run MD simulations with PLUMED.

The simulation with pmemd.cuda failed just in seconds after reading the PLUMED input file.

The problem is:

PLUMED:   on file colvar.dat
PLUMED:   with format  %f
PLUMED: END FILE: plumed.dat
pmemd.cuda: Plumed.h:2605: plumed_gcreate: Assertion `plumed_gmain.p==((void *)0)' failed.

The PLUMED input is very simple, just writing out a distance:

##

 dist: DISTANCE ATOMS=1,3817

PRINT STRIDE=500 ARG=dist FILE=colvar.dat

The simulation was OK if I turned PLUMED off (plumed=0).

Does anyone have the same problem? Thanks a lot!

All the best,

Qinghua

[viktor drobot]

Hello! Could you please provide us with full specs of these clusters? I mean compiler versions, detailed configuration and building protocol of Amber and PLUMED, kernel versions, hardware specs...

[Qinghua Liao]

Hello Viktor,

Thanks a lot for your prompt response!

The one which works is on Ubuntu 16.04 (Linux version 4.15.0-91-generic (gcc version 5.4.0 20160609 (Ubuntu 5.4.0-6ubuntu1~16.04.12))),

The compilers are GCC/8.3.0 (GFORTRAN)  CUDA/10.1.243  OpenMPI/3.1.4  (GPU: K80, V100)

The one which does not work is on openSUSE Leap 42.2 (Linux version 4.4.27-2-default (gcc version 4.8.5 (SUSE Linux) ))

The compilers are GCC/4.8.5 (SUSE Linux), OpenMPI/1.10.3, CUDA/9.0 (GPU: Geforce1080 TI)

The installation protocol is the same on both clusters:

git clone https://github.com/plumed/plumed2.git

cd plumed2

./configure --prefix=/path/

source sourceme.sh

cd amber18

export AMBERHOME=/path/amber18

plumed-patch -p (choose amber 18)

./configure -cuda gnu

make install -j 8

source amber.sh

I tried sander on the not-working cluster, it worked.

I guess there might be something wrong with the compilation.

Thanks for looking into it.

All the best,

Qinghua

[viktor drobot]

If I remember correctly, here was a thread some months ago dealing with the similar problem. And if I remember it right the cause was old gcc compiler from 4.x branch. There was a lot of innovations regarding C++11 support in versions 5.x and higher so may be it worth to try newer compiler, if possible. Do you have access for another compilers on problematic cluster?

[viktor drobot]

yes, I've found it. OP didn't tell his compiler info, but error is the same. I assume it's all about C++ features support in older GCCs. May be PLUMED developers could say if there any special requirements for C++11 standard...

[Qinghua Liao]

git clone https://github.com/plumed/plumed2.git

cd plumed2

./configure --prefix=/path/

source sourceme.sh

 

Here, I forgot to mention "make -j 8" and "make install -j 8" after configuration.

[Qinghua Liao]

Thanks a lot, Viktor!

It's our own cluster, we will try to install new compilers, and then try it again!

All the best,

Qinghua

[Qinghua Liao]

Hello Viktor,

I tried GCC-6.5.0 and GCC-8.4.0, I used the new compiler to compile PLUMED and openMPI.

The problem is still the same! Any clue? Thanks!

All the best,

Qinghua

On Tuesday, March 24, 2020 at 12:30:52 AM UTC+1, viktor drobot wrote:

[viktor drobot]

The only things that remains is the Linux kernel version. Is there any possibility to update it?

[Qinghua Liao]

Thanks Viktor!

I am not going to do that, as I worry about that I mess up the cluster.

I will run simulations on the other cluster.

All the best,

Qinghua