The ABINIT Users Mailing List ( CLOSED )

[Xavier Gonze <gonze@pcpm.ucl.ac.be>]

Paul Fons wrote:
> Hello.  I have what I hope is a simple question.  I wish to run abinit 
> to check on the electronic properties of some of the model structures we 
> have been developing.  To this end, I have (just) purchased at 2.7 GHz 
> dual G5 with 4 GB of memory and have ordered the absoft compiler.  Once 
> I have the compiler in hand, I plan to compile the latest version of 
> abinit (as there seem to be no binaries available).  My question is 
> before I start compiling -- I have two cpu's, will I get any more speed 
> by compiling the MPI version (and installing the appropriate MPI 
> subsystem) or will the serial version give roughly the same performance 
> for my setup.  I realize the answer probably depends on the details of 
> the calculation, I am really asking about typical and best case 
> results.  Thanks for any help.
>
>                                                                     Paul 
> Fons

Dear Paul Fons,

In most cases, you will be able to get nearly a factor of two speed up
with your bi-proc :
- if you have 2k points or more, the k point parallelisation is OK,
 and very easy to use
- even with 1k point, you can use spin parallelisation if your material
 is spin polarized, or band parallelisation. The latter is a bit more
 difficult to use, and is restricted to a small number of procs (2 is OK).
Please see the discussion in ~ABINIT/Infos/paral_use ,
as well as the description of the input variable wfoptalg .

Xavier