[Pw_forum] Low speed

[mohsen modaresi]

PS,
We increased speed with install some external libraries (BLAS, LAPACK,
and MPI Lib.), I hope it help to other users.
Mohsen


On 5/20/10, mohsen modaresi <modaresi.mohsen at gmail.com> wrote:
> Dear Baris
> Thanks a lot for your answer.
>  how many scf steps
> "do you perform in these 3 hours? Does your system have some
> convergence problems?", It take 12 iteration to perform, And there is
> no convergence problem
> "This is especially important if you are using a low memory
> bandwidth architecture.", I can not understand your hint, We use a P4
> machine with 8 core, I think you say we must use Half of them, Is it
> true?
> And finally thanks for your answer
> Moshen Modaresi
>
>
> On 5/20/10, O. Baris Malcioglu <baris.malcioglu at gmail.com> wrote:
>> Dear Mohsen,
>> Please notice that you have asked a rather vague question regarding
>> performance,
>>
>> First of all, please notice that since PW is a plane wave code, it is
>> not just the number of atoms that determine the "size" of the system.
>> You have to take into account the number of plane waves you are using.
>> This depends on your cutoffs and simulation supercell. Exaggerating
>> the situation unnecessarily, for example putting only an ethane
>> molecule in a 500x500x500 Bohr box (even with modest cutoff) will
>> result in a system size that will certainly give a hard time to the
>> computer you are describing. In the same sense, using very high
>> cutoffs will lead to same problem.
>>
>> Secondly, "3 hours" is not a good enough criteria, how many scf steps
>> do you perform in these 3 hours? Does your system have some
>> convergence problems?
>>
>> Finally, 8 cpu is not a good enough identification nowadays, if the
>> "cpu" you are mentioning is due to hyper-threading, it is usually
>> better if you use only 4, since you waste your precious memory
>> bandwidth. This is especially important if you are using a low memory
>> bandwidth architecture.
>>
>> Some general guidelines :
>>
>> -Make sure that your computer is not using swap,
>> -Use architecture dependent acceleration libraries such as MKL. For
>> example you can setup fftw 3.x to utilize instruction sets such as
>> sse2.
>> -If this is a parallel compiler check wall times, how much do you
>> loose communicating?
>>
>> Best,
>> Baris
>>
>> 2010/5/20 Phillip Nyawere <pnyawere at gmail.com>:
>>> You need high speed computer. I tried with 96 atoms and it was taking
>>> even
>>> longer.
>>> Success.
>>>
>>> On Wed, May 12, 2010 at 10:07 PM, mohsen modaresi
>>> <modaresi.mohsen at gmail.com> wrote:
>>>>
>>>> Dear Developers and users,
>>>> I run a SCF calculation with 71 atoms on a camputer with 8 cpu and 8GB
>>>> RAM, i expect it run in some minutes, But it take 3 hours to run.
>>>> Is it regular?
>>>>
>>>>
>>>> _______________________________________________
>>>> Pw_forum mailing list
>>>> Pw_forum at pwscf.org
>>>> http://www.democritos.it/mailman/listinfo/pw_forum
>>>>
>>>
>>>
>>>
>>> --
>>> Phillip W. Otieno Nyawere,
>>> Kabarak University,
>>> Dept of Physics & Mathematics,
>>> P.O Box Private Bag - 20157,
>>> Kabarak, Kenya.
>>> Tel +254728342054
>>> pnyawere at gmail.com, potieno at kabarak.ac.ke
>>>
>>> The battle belongs to the Lord.
>>>
>>> _______________________________________________
>>> Pw_forum mailing list
>>> Pw_forum at pwscf.org
>>> http://www.democritos.it/mailman/listinfo/pw_forum
>>>
>>>
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://www.democritos.it/mailman/listinfo/pw_forum
>>
>