Re: CUDA vd non-CUDA for minimization

From: Francesco Pietra (chiendarret_at_gmail.com)
Date: Sat Oct 19 2013 - 02:28:50 CDT

>
> Search the archives (preferably before posting)
>

Good for you if you find the time to search among archives (as these are
organized). At any time, I was aware (as I mentioned) of the problem from
personal past experience. And I proposed the same alternative to minimize
that I proposed in the past.

If you read better, my message was to remark that the situation needs
attention, not as far as minimization is concerned (for which I have an
alternative) but from what in general comes out from graphic cards. Given
the situation, when energy problems are involved, I continue on the BluGene
instead of faster mainframes bases on graphic cards.

best regards
francesco

On Fri, Oct 18, 2013 at 8:55 PM, Kenno Vanommeslaeghe <
kvanomme_at_rx.umaryland.edu> wrote:

> This has been discussed in great detail by Axel Kohlmeyer in a thread
> titled "Equilibration of the membrane bilayers with CUDA-support" only 5
> weeks ago. Search the archives (preferably before posting).
>
> To summarize: the CUDA code is expected to run into trouble when given
> strained structures that won't occur in a normal MD simulation.
>
>
>
> On 10/18/2013 01:13 PM, Francesco Pietra wrote:
>
>> Hello:
>> With GTX-680 cards and cuda namd 2.9 I had problems for minimization,
>> problems that I already encountered with other systems and previous
>> versions of cuda namd.
>>
>>
>> I refer to a difficult situation, where i had rotated a histidine residue
>> in the protein and imposed rmsd colvars. With cuda namd only
>> velocityQuenching could be run, while "minimization" resulted in atoms
>> moving too fast (just in the area that I had modified) under a variety of
>> settings.
>>
>> Turning to non cuda namd 2.9 "minimization" went on without problems. It
>> seems that when the calculation of energy is most relevant, as during a
>> minimization, CPUs are still a better choice.
>>
>>
>> If true, I wonder whether free energy calculations with cuda are still to
>> be taken with much reservation.
>>
>> thanks
>> francesco pietra
>>
>>
>

This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:21:47 CST