Re: [AMBER] serial compilation of pmemd

From: <>
Date: Mon, 22 Mar 2010 15:51:02 +0200

I have run "mpirun -np 1 pmemd ..." for our pmemd9 which is supposed
to have been compiled in parallel. It runs without complaining. So,
according to Dr. Case's suggestion below, it seems that still the
serial version is running for some unknown reason. However, people at
our high performance computing center say that the shared library
dependencies of the pmemd binary file (see below) indicate that it is
the parallel version.
hsp70.bulent.d001 1node]$ which pmemd
[hsp70.bulent.d001 1node]$ ldd /RS/progs/amber9/exe/pmemd => /lib64/ (0x0000003641200000) =>
(0x00002b235f6a9000) =>
(0x00002b235fa3b000) => /RS/progs/intel/impi/3.1/lib64/
(0x00002b235fc52000) =>
/RS/progs/intel/impi/3.1/lib64/ (0x00002b235ff78000) =>
(0x00002b23600a7000) => /lib64/ (0x000000363fa00000) => /lib64/ (0x0000003640a00000) => /lib64/ (0x0000003640e00000) => /lib64/ (0x0000003640600000) => /lib64/ (0x00000039b5c00000)
         /lib64/ (0x000000363f600000)
So, I am a bit confused whether I am running the serial or parallel
version of pmemd. Any suggestion will be appreciated. Thanks in advance.

Alinti case <>

> On Thu, Feb 25, 2010, wrote:
>> Pmemd has been successfully compiled only in serial. Parallel
>> compilation appears to be successful, but when we try to run in
>> parallel, there is no speed up with respect to a single processor run.
>> When I look at the job status in the queue, I see that it runs on
>> multiple processors, but the speed is the same as a single processor
>> run. Also, in the output file the number of processors used is not
>> written (I guess, for parallel pmemd, it should be written somewhere
>> near the beginning of the output file).
> My best (wild) guess is that your mpirun is still pointing to the serial
> version of pmemd. Unlike sander, which creates separate executables
> for serial and parallel ("sander" and "sander.MPI"), the pmemd setup uses
> the same executable name for both serial and parallel versions; the idea
> is that the "nopar" version is hardly ever used in practice.
> Your symptoms (several processors running, but not having the number of
> processors in the output) suggests that you are still running the serial
> version. Check your timestamps carefully. Another trick is to run
> "mpirun -np 1 pmemd ....". The serial pmemd would be OK with this, but the
> parallel version will complain that it cannot run on one processor. So this
> can be used to make sure you are actually running the parallel version.
> Of course, the error could be somewhere else completely, but the above is
> worth checking.
> ...dac
> _______________________________________________
> AMBER mailing list

AMBER mailing list
Received on Mon Mar 22 2010 - 07:00:03 PDT
Custom Search