[Pw_forum] Intel MPI

[joaquin peralta]

Hello,

On Wed, Aug 18, 2010 at 12:28 PM, Paolo Giannozzi wrote:

>
> On Aug 18, 2010, at 19:03 , joaquin peralta wrote:
>
> > We use a surface of Al with 81 atoms in 128 cpus using intelMPI and
> > openMPI 1.4
>
> but not in the same way! the first job uses 8 pools of 16 processors,
> the second
>

Yes it's true, not the same way.


> 1 pool of 128 processors. The first choice seems to be reasonable,
> the second
>

After i recheck my times again, and compare a more realistic run.


> isn't, since an enormous amount of time is spent in communications
> (look at the
> time spent in fft_scatter).
>
>
Is possible know previously how is the best parameter to use with -np, I
mean if somebody recommend me a thread about the uses (i don't understand
very well the manual yet) of the optimization flags (pool, image, ntg,
diag).

Thanks in advance

Joaquin Peralta


> P.
> ---
> Paolo Giannozzi, Dept of Physics, University of Udine
> via delle Scienze 208, 33100 Udine, Italy
> Phone +39-0432-558216, fax +39-0432-558222
>
>
>
> ___
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>



-- 

Group of NanoMaterials

http://www.gnm.cl

Joaqu?n Andr?s Peralta Camposano

http://zeth.ciencias.uchile.cl/~jperalta

In a world without frontiers,
who needs Gates and Win.
-- next part --
An HTML attachment was scrubbed...
URL: 
http://www.democritos.it/pipermail/pw_forum/attachments/20100819/742ae77c/attachment-0001.htm
 

[Pw_forum] Intel MPI

[Paolo Giannozzi]

On Aug 18, 2010, at 19:03 , joaquin peralta wrote:

> We use a surface of Al with 81 atoms in 128 cpus using intelMPI and  
> openMPI 1.4

but not in the same way! the first job uses 8 pools of 16 processors,  
the second
1 pool of 128 processors. The first choice seems to be reasonable,  
the second
isn't, since an enormous amount of time is spent in communications  
(look at the
time spent in fft_scatter).

P.
---
Paolo Giannozzi, Dept of Physics, University of Udine
via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222

[Pw_forum] Intel MPI

[Goranka Bilalbegovic]

On Wed, Aug 18, 2010 at 3:16 PM, Paolo Giannozzi wrote:
Can anybody report any experience linking with Intel version of MPI?
---

Intel MPI is installed on the cluster I am using. Several recent versions of
QE (including 4.2.1) are compiling and running without any
problems.

However, I do have problems in running SaX jobs on the same cluster. The
problems with running SaX jobs started last year when this cluster was
changed and several new libraries and compilers were installed at the same
time. Together with the system manager I was trying to detect the problem
with SaX. The frst idea was that this is because of (then new for us) the
Lustre file system, the second was mvapich.  A new version of mvapich was
installed, but it did not help. There is a possibility that SaX jobs do not
run on this cluster (they stayed in the queue forever) because of the
problems with Intel MPI.

Best regards,
-- 
Goranka Bilalbegovic,
Department of Physics, Faculty of Science,
University of Zagreb, Croatia
-- next part --
An HTML attachment was scrubbed...
URL: 
http://www.democritos.it/pipermail/pw_forum/attachments/20100818/be7bd64e/attachment.htm
 

[Pw_forum] Intel MPI

[Paolo Giannozzi]

Can anybody report any experience linking with Intel
version of MPI?

P.
---
Paolo Giannozzi, Dept of Physics, University of Udine
via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222

[Pw_forum] Intel MPI

[joaquin peralta]

Me and Claudia

are using intelmpi 4.0 in http://cystorm.ece.iastate.edu with a very good
performance .. but we are not sure now ... let me explain :

We use a surface of Al with 81 in 128 cpus using intelMPI and openMPI 1.4

The compilations options (make.sys), input file and output files are in this
page (sorry for use my page but was the more fast) :

http://www.lpmd.cl/jperalta/index.php?n=Site.QE-Cystorm

Probably is not a good test, because we detect after that the cluster
schedule manager (torque) was not work properly with intelMPI and was very
probably that some external works are using process in the open-mpi running.
Look this thread about schedule-manager and intelMPI
http://software.intel.com/en-us/forums/showthread.php?t=76537

Right now the cluster is under a upgrade, sure when we can do more test we
shared that.

Best Regards
Joaquin and Claudia.


On Wed, Aug 18, 2010 at 8:16 AM, Paolo Giannozzi wrote:

> Can anybody report any experience linking with Intel
> version of MPI?
>
> P.
> ---
> Paolo Giannozzi, Dept of Physics, University of Udine
> via delle Scienze 208, 33100 Udine, Italy
> Phone +39-0432-558216, fax +39-0432-558222
>
>
>
> ___
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
>



-- 

Group of NanoMaterials

http://www.gnm.cl

Joaqu?n Andr?s Peralta Camposano

http://zeth.ciencias.uchile.cl/~jperalta

In a world without frontiers,
who needs Gates and Win.
-- next part --
An HTML attachment was scrubbed...
URL: 
http://www.democritos.it/pipermail/pw_forum/attachments/20100818/4637f898/attachment.htm
 

[Pw_forum] Intel MPI

[Lazaro Calderin]

We have been using mpiifort and mpiicc without major issues. If my
memory serves me well they use the intel mpi libs while
mpif90 and the rest of the gnu compilers provided by intel use mpich.
Regards, L.

On Wed, 2010-08-18 at 15:16 +0200, Paolo Giannozzi wrote:
> Can anybody report any experience linking with Intel
> version of MPI?
> 
> P.
> ---
> Paolo Giannozzi, Dept of Physics, University of Udine
> via delle Scienze 208, 33100 Udine, Italy
> Phone +39-0432-558216, fax +39-0432-558222
> 
> 
> 
> ___
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
> 
-- 
Lazaro Calderin
Department of Physics
Queen's University
Kingston, Ontario
Canada

[Pw_forum] Intel MPI or OpenMPI?

[Nicola Marzari]

Dear Tran,

here are the Q-E tests for CP in a Q6600 :
http://quasiamore.mit.edu/pmwiki/index.php?n=Main.CP90Timings

If you get the same performance, things are running ok.

Always set the system varable OMP_NUM_THREADS to 1, otherwise
perfromance can easily decrease by one order of magnitude, on a 4
core job.

nicola



Ha, Tran Van wrote:
> Thanks Nicola Marzari  and  Jun Dai for help me about FFTW.
> Dear Pwscf Users, I have a problem with my new quadcore Q6600 (I had no 
> problem with my old parrallel machine).
> I read the messages from our forum, especially 
> http://www.democritos.it/pipermail/pw_forum/2008-February/subject.html#8280
> As *Axel Kohlmeyer* said, we can get a speedup of roughly 2.5x to 3.5x from
> a quad core cpu over a corresponding single core but to me it's very 
> bad! using one core always run faster all 4 cores (the same with gamma 
> point also k-point).
> I used PWscf 4.0.3, Intel compiler 10.1.018 and OpenMPI 1.2.8. I don't 
> know why? Or I can change into using Intel MPI with my Quadcore?
> Thank you in advance for your help.
> 
> 
> Nam, Tran Van
> Hanoi University of Technology
> Faculty of Chemistry
> Department of Physical Chemistry
> 
> 
> 
> Get your preferred Email name! 
> 
>  
> 
> Now you can @ymail.com and @rocketmail.com.
> 
> 
> 
> 
> ___
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum


-- 
-
Prof Nicola Marzari   Department of Materials Science and Engineering
13-5066   MIT   77 Massachusetts Avenue   Cambridge MA 02139-4307 USA
tel 617.4522758 fax 2586534 marzari at mit.edu http://quasiamore.mit.edu

[Pw_forum] Intel MPI or OpenMPI?

[Ha, Tran Van]

Thanks Nicola Marzari??and? Jun Dai for help me about FFTW.
Dear Pwscf Users, I have a problem with my new quadcore Q6600 (I had no problem 
with my old parrallel machine).
I read the messages from our forum, especially 
http://www.democritos.it/pipermail/pw_forum/2008-February/subject.html#8280
As Axel Kohlmeyer said,?we can get a speedup of roughly 2.5x to 3.5x from
a quad core cpu over a corresponding single core but to me it's very bad! using 
one core always run faster all 4 cores (the same with gamma point?also k-point).
I used PWscf 4.0.3, Intel compiler 10.1.018 and OpenMPI 1.2.8. I don't know 
why? Or I can change into using Intel MPI with my Quadcore?
Thank you in advance for your help.


Nam, Tran Van
Hanoi University of Technology
Faculty of Chemistry
Department of Physical Chemistry


  Get your new Email address!
Grab the Email name youve always wanted before someone else does!
http://mail.promotions.yahoo.com/newdomains/aa/
-- next part --
An HTML attachment was scrubbed...
URL: 
http://www.democritos.it/pipermail/pw_forum/attachments/20081023/bcd5cc65/attachment.htm