[CPMD-list] MEMORY ALLOCATION FAILED at Cray XT3

[Ari P Seitsonen]

Dear Alexandr,

   Adding to Alex's [Kohlmeyer] answer, it is memory-wise cheaper to use 
PCG [MINIMIZE] than the default ODIIS, as the former only needs to store 
the whole set of wave functions twice whereas the default for the latter 
is to store ten vectors (this can be usually reduced without too big a 
loss in the convergence rate, keyword 'ODIIS' and the number of vectors 
on the following line).

     Greetings from the rainy Paris,

        apsi

-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-=*=-
   Ari P Seitsonen / Ari.P.Seitsonen at iki.fi / http://www.iki.fi/~apsi/
   CNRS & IMPMC, Université Pierre et Marie Curie
   Tel: +33-1-4427 7542, Fax: +33-1-4427 3785, GSM: +33-6-6736 3820

On Wed, 23 Aug 2006, Alexandr Isayev wrote:

> Dear CPMD community:
>
> I experienced strange cpmd (v3.11.1) behavior at Cray XT3.
> *ANY* job that requires more than few hundred Mb of memory fails with
>
> PROGRAM STOPS IN SUBROUTINE MEMORY| ALLOCATION FAILED (PME) [PROC=   0]
>
> However,this particular XT3 has 2gb per CPU. First, I thought that yod
> can't allocate required amount of memory by default. I played with
> different setting, etc; it does not change anything. I also worked with
> local admins. They tested and found no problems with yod, catamount
> nodes, etc. It can allocate more than 1.8G with default settings.
> I also recompiled the code with default XT3 settings, but it did not
> help either.
>
> This particular example below, just a standard box with 216 waters,
> PBE, 80Ry cutoff. It needs about 1Gb per CPU or less.
>
> Can someone confirms my observations with other XT3s?
>
> Thank you in advance,
> Alexandr
>
> Relevant output part is attached below:
> ==================================================================
>
> PARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARA
>  NCPU     NGW     NHG  PLANES  GXRAYS  HXRAYS ORBITALS Z-PLANES
>     0   16665  133357      13     498    1986      54       1
>     1   16669  133355      14     498    1986      54       1
>     2   16669  133337      13     498    1986      54       1
>     3   16669  133333      14     498    1986      54       1
>     4   16669  133349      13     498    1986      54       1
>     5   16667  133340      14     498    1988      54       1
>     6   16665  133340      13     498    1988      54       1
>     7   16669  133336      14     498    1988      54       1
>     8   16669  133244      13     497    1987      54       1
>     9   16669  133318      14     498    1988      54       1
>    10   16669  133332      13     498    1988      54       1
>    11   16663  133332      14     498    1988      54       1
>    12   16661  133338      13     498    1988      54       1
>    13   16661  133324      14     498    1988      54       1
>    14   16659  133318      13     498    1988      54       1
>    15   16656  133322      14     496    1988      54       1
>                G=0 COMPONENT ON PROCESSOR :     8
> PARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARAPARA
>
> ***    LOADPA|  CURRENT HEAP USED/FREE    86007/ 1824776 kBytes ***
>
> OPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPEN
> NUMBER OF CPUS PER TASK                                        1
> OPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPENMPOPEN
>
> ***     RGGEN|  CURRENT HEAP USED/FREE    90791/ 1819992 kBytes ***
>
> ************************** SUPERCELL ***************************
> SYMMETRY:                                           SIMPLE CUBIC
> LATTICE CONSTANT(a.u.):                                 35.34045
> CELL DIMENSION:  35.3404  1.0000  1.0000  0.0000  0.0000  0.0000
> VOLUME(OMEGA IN BOHR^3):                             44138.34843
> LATTICE VECTOR A1(BOHR):           35.3404     0.0000     0.0000
> LATTICE VECTOR A2(BOHR):            0.0000    35.3404     0.0000
> LATTICE VECTOR A3(BOHR):            0.0000     0.0000    35.3404
> RECIP. LAT. VEC. B1(2Pi/BOHR):      0.0283     0.0000     0.0000
> RECIP. LAT. VEC. B2(2Pi/BOHR):      0.0000     0.0283     0.0000
> RECIP. LAT. VEC. B3(2Pi/BOHR):      0.0000     0.0000     0.0283
> REAL SPACE MESH:                   216          216          216
> WAVEFUNCTION CUTOFF(RYDBERG):                           80.00000
> DENSITY CUTOFF(RYDBERG):          (DUAL= 4.00)         320.00000
> NUMBER OF PLANE WAVES FOR WAVEFUNCTION CUTOFF:            266649
> NUMBER OF PLANE WAVES FOR DENSITY CUTOFF:                2133275
> ****************************************************************
>
> ***  RINFORCE|  CURRENT HEAP USED/FREE    97593/ 1813190 kBytes ***
> ***    FFTPRP|  CURRENT HEAP USED/FREE   116984/ 1793799 kBytes ***
>
> GENERATE ATOMIC BASIS SET
>      O        SLATER ORBITALS
>        2S        ALPHA=   2.2458      OCCUPATION= 2.00
>        2P        ALPHA=   2.2266      OCCUPATION= 4.00
>      H        SLATER ORBITALS
>        1S        ALPHA=   1.0000      OCCUPATION= 1.00
>

...

>
> -------------------------------------------------------
> Alexandr Isayev,
> Graduate Research Assistant, and System Administrator
> @ Computational Center for Molecular Structure
> and Interactions (CCMSI),
> Jackson State University,
> Jackson, MS USA
>   Tel:  +(601) 979-1134
> e-mail:  alex(at)ccmsi.us
>   Web:  http://www.ccmsi.us
> --------------------------------------------------------
>
> _______________________________________________
> CPMD-list mailing list
> CPMD-list at cpmd.org
> http://cpmd.org/mailman/listinfo/cpmd-list
>