[gmx-users] MVAPICH2 mdrun problem for SD and MD, GROMACS 4.0.5

[Mark Abraham]

Daniel Adriano Silva M wrote:
> Mark,
> 
> Again me. I reinstalled things, you must be thinking that I am wrong
> trying to get the icc installation to work, but as I stated on a
> previous mail I get very good scaling and over 40% better performance
> with icc than gcc. I had tested gmx_test, and all appears to be ok,
> but I think the gmx test is incomplete for mpi, since some test does
> not run with dd and others with pd. Please, are there any chance that

Yes, the test set is not useful for parallel tests, since many of the 
systems are too small to decompose.

> you have some tprs that could work as a test-set, with results that
> can be compared????

You can roll your own, here... Compile with CFLAGS=-O0, the built-in 
fftpack, blas and lapack, and use environment variable NOASSEMBLYLOOPS=1 
to get a bare-bones very-likely-correct version. Compare output .edr 
files using gmxcheck with whatever you want to verify.

Mark

> Thank you
> Daniel
> 
> 2009/10/30 Mark Abraham <Mark.Abraham at anu.edu.au>:
>> Daniel Adriano Silva M wrote:
>>> Mark,
>>>
>>> Then I will try with other versions, check gmx-tests and not mix
>>> compilers, thank you much by your help. Last question, I am afraid to
>>> ask, but I have some simulations that were ran with this setup,
>>> energy, temp, and pressure looks good on all, and it ran by pretty
>>> long time (several ns) without any apparent errors, do you think that
>>> there could be any way to check if these md-sims results are actually
>>> wrong?
>> That's a hard question, because MD with even the same GROMACS version with
>> different compilers can diverge rapidly. If you still had the old version,
>> you might try a new mdrun -reprod with nstlog=1 on both versions. If they
>> give the same output, you might then have some confidence that the old
>> version was only wrong under some conditions, and hopefully all of those
>> provoked crashes. However it would be likely that they wouldn't give the
>> same output even if both were functioning correctly. Still, the energy
>> differences over the first few MD steps should be down at about the
>> precision reported in the .log file, and that's useful knowledge.
>>
>> Mark
>> _______________________________________________
>> gmx-users mailing list    gmx-users at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-users
>> Please search the archive at http://www.gromacs.org/search before posting!
>> Please don't post (un)subscribe requests to the list. Use the www interface
>> or send it to gmx-users-request at gromacs.org.
>> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
>>
> _______________________________________________
> gmx-users mailing list    gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> Please search the archive at http://www.gromacs.org/search before posting!
> Please don't post (un)subscribe requests to the list. Use the 
> www interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
>