Re: Constraint failure in RATTLE

From: Massimiliano Porrini (mozz76_at_gmail.com)
Date: Mon Mar 31 2014 - 11:48:38 CDT

On 31 March 2014 07:43, Norman Geist <norman.geist_at_uni-greifswald.de> wrote:

> Just as a additional hint, memory errors can cause such a behavior.
> Especially broken GPUs. So you might have a look on specific nodes being
> present while the error occurs and not being present if the error is absent.
>

Good point and actually I had already considered this explanation, but then
I gave a try
with pmemd.cuda.MPI of Amber12 and it worked very smoothly for "rather
long" runs:
25,000,000 steps with no interruption (time-step = 2 fs).

By the way, given that CUDA-enabled versions of NAMD and AMBER work
differently
one from another, I am not that technologically expert to claim that my
GPUs are not broken...
Nevertheless, one code gives problems while the other one does not.

Thanks,
Max

>
>
> Norman Geist.
>

Cheers,
Massimiliano

>
>
> *Von:* owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] *Im
> Auftrag von *Massimiliano Porrini
> *Gesendet:* Sonntag, 30. März 2014 13:50
> *An:* Francesco Pietra
> *Cc:* NAMD
> *Betreff:* Re: namd-l: Constraint failure in RATTLE
>
>
>
> Hi,
>
>
>
> I had been experiencing this problem rather frequently in my 2 types of
> simulation:
>
>
>
>
>
> 1) With namd2.9 - CUDA enabled CVS-2013-07-22_Linux-x86_64-multicore .
>
>
>
> Each time the RATTLE algorithm failed for a different carbon atom, but
>
> always the same type, CTL2, which is a carbon atom of an aliphatic tail
>
> of a amphiphile molecule.
>
>
>
> I am using CHARMM36 force field parameters.
>
>
>
>
>
> 2) The second has happened within another compound, but with either
>
> an aromatic carbon, CA, or an sp3 aliphatic carbon, CG331, bound to the
> benzene ring.
>
>
>
> In this case, the failure was encountered with both the CUDA enabled and
> lone CPU version
>
> (2.9 and the following CVS: 2013-11-09, 2014-01-13 and 2014-03-23).
>
>
>
> However this compound was parametrised by me with the ffTK of VMD, hence
>
> I have always attributed the failure to an non-optimal modellisation.
>
>
>
> In this case, I have also tried to run the simulation with the following
> "slower" and more accurate
>
> settings (after the system total energy was sufficiently minimised: it has
> plateaued for tens of thousands
>
> of steps..!):
>
>
>
>
>
> timestep 1.0
> rigidBonds all
> nonbondedFreq 1
> fullElectFrequency 1
> stepspercycle 10
>
>
>
>
> But unfortunately no luck!
>
>
>
> Regards,
>
> Massimiliano
>
>
>
>
>
> PS in all the cases I have just restarted the simulation and reached
>
> the desired simulation time (by the way, I want to point out that
>
> for some simulation portions no RATTLE failure occurred).
>
>
>
>
>
>
>
>
>
>
>
> On 30 March 2014 09:31, Francesco Pietra <chiendarret_at_gmail.com> wrote:
>
> Hello:
>
> This is to report a rather unusual "constraint failure in RATTLE algorithm
> for atom 520" during non-periodic accelMD under GB conditions with namd2.9.
>
> Atom 520 is CA of a TRP residue that makes part seven residues at one end
> of a 37-residue peptide, and which are under rmsd colvars, keeping them as
> an alpha-helix stretch.
>
> This accelMD followed a series of similar accelMD, which were started
> after a series of T-REMD under either non-periodic GB or periodic explicit
> conditions. There was no crash before the one above. Otherwise, folding was
> going on smoothly, albeit still far from completion.
>
> Graphics does not suggest any unusual situation for CA 520. Also, namd
> plot with the log file shows constant energy, even when magnified.
>
>
>
> francesco pietra
>
>
>
>
>
> --
> Dr Massimiliano Porrini
> Institut Européen de Chimie et Biologie (IECB)
> CNRS UMR 5248 Chimie et Biologie des Membranes et des Nano-objets (CBMN)
> 2, rue Robert Escarpit
> 33607 Pessac Cedex
> France
> http://www.iecb.u-bordeaux.fr/
>
> Emails: m.porrini_at_iecb.u-bordeaux.fr
> M.porrini_at_ed.ac.uk
> maxp_at_iesl.forth.gr
> mozz76_at_gmail.com
>
>
> ------------------------------
> <http://www.avast.com/>
>
> Diese E-Mail ist frei von Viren und Malware, denn der avast! Antivirus<http://www.avast.com/>Schutz ist aktiv.
>
> --
> Dr Massimiliano Porrini
> Institut Européen de Chimie et Biologie (IECB)
> CNRS UMR 5248 Chimie et Biologie des Membranes et des Nano-objets (CBMN)
> 2, rue Robert Escarpit
> 33607 Pessac Cedex
> France
> http://www.iecb.u-bordeaux.fr/
>
> Emails: m.porrini_at_iecb.u-bordeaux.fr
> M.porrini_at_ed.ac.uk
> maxp_at_iesl.forth.gr
> mozz76_at_gmail.com
>

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:22:16 CST