[Getdp] Usage of multicore CPUs

[Christophe Geuzaine]

On 23/09/10 08:48, Lars Rindorf wrote:
> Hi Michael
>
> Yes, it is possible. If for example you have an Intel CPU then you can buy intel mkl blas. If you compile getdp from source with this blas then it will be parallelized on the CPU for certain matrix solvers. The matrix assembly will remain sequential despite using multi core blas. This is due to the construction of getdp.
>
> To run getdp on a cluster then you need to partition the matrix. Gmsh can do this but getdp is not currently able to this. I think Christophe is planning to do incorporate it in some future version.
>

Hi Guys - In addition to multi-threaded BLAS, you can now also use getdp 
fairly efficiently for parallel linear system solves on distributed 
memory machines.

The latest SVN snapshots incorporate various fixes and optimizations so 
that getdp can use all the PETSc solvers using MPI in parallel. Matrix 
assembly will not be sped up (and will never be; but we are currently 
developing GetDP's fully-parallelized successor) but the matrix is 
distributed across the CPUs and the solution is fully parallelized. 
Post-processing must be done sequentially, but has been optimized with 
support for the new Gmsh mesh-based format so it can be run on 
arbitrarily large datasets.

Typical use in parallel for large problems is thus:

   mpirun -np 16 getdp myfile.pro -solve myresolution
   getdp myfile.pro -pos -bin -v2

This has been successfully tested with both iterative (GMRES+SOR) and 
direct (MUMPS) parallel solvers on up to 128 CPUs, for test-cases up to 
10 millions dofs.

Christophe



> KR
> Lars
>
>
>
> -----Oprindelig meddelelse-----
> Fra: getdp-bounces at ace20.montefiore.ulg.ac.be [mailto:getdp-bounces at ace20.montefiore.ulg.ac.be] På vegne af michael.asam at infineon.com
> Sendt: 10. september 2010 15:42
> Til: getdp at geuz.org
> Emne: [Getdp] Usage of multicore CPUs
>
> Hi all,
>
> does anyone know if it is possible to use more than one CPU core with GetDP in order to speed up the simulation?
> Thank's for your help!
>
> Best regards,
> Michael Asam
>
>
>
> _______________________________________________
> getdp mailing list
> getdp at geuz.org
> http://www.geuz.org/mailman/listinfo/getdp
>
> _______________________________________________
> getdp mailing list
> getdp at geuz.org
> http://www.geuz.org/mailman/listinfo/getdp
>


-- 
Prof. Christophe Geuzaine
University of Liege, Electrical Engineering and Computer Science
http://www.montefiore.ulg.ac.be/~geuzaine