The ABINIT Users Mailing List ( CLOSED )

["Sam M. Vinko" <Sam.Vinko@physics.ox.ac.uk>]

Hello,

I have a question regarding the way abinit splits up cpu resources accros the kpt/band/fft levels of parallelization (on abinit 5.8.4).

My calculation is a 32 atom Al supercell, looking at 64 bands, and I'd like to sample the BZ with a 8x8x8 MP grid which, as I've excluded all symmetries, contains 256kpt. 

I have 128 cpus for this calculation and I'd like to distribute resources in the most efficient way so I've used the input option PARAL_KGB -128 which should tell me all the possible ways of splitting the cpus between bands/ffts/kpts; however, I seem to only get one option:

For dataset=   1  a possible choice for less than  128 processors is:

  nproc    npkpt    npband     npfft    bandpp

   256     256        1         1          2

Indeed, it turns out that any way of distributing 128 cpus for this job crashes.

Does this mean that you always need at least a number of cpus larger than the k-grid? Or is there some subtlety I am missing?

Sample input file with only the relevant option (takes about a minute to run) is appended below.

Thanks in advance for your answers,

Sam  

#*****************************************************************************

#Definition of parallelizations

paral_kgb -128

#npband 32

#npfft 1

#npkpt 4

#Definition of occupation numbers

occopt 3               #FD distribution

tsmear 0.003

#Definition of the unit cell

acell 3*15.3271402862  # 7.6635701431 per cell, from minimizing forces

rprim  1.0  0.0  0.0

       0.0  1.0  0.0 

       0.0  0.0  1.0

#Other optimizations

optcell 0

wfoptalg 4

fftalg 401 

fft_opt_lob 2 

nloalg 4

iscf 17

#Definition of the atom types

ntypat 1          

znucl 13        

#Definition of the atoms

natom 32    

typat 32*1       

xred   0.00  0.00  0.00

       0.50  0.50  0.50

       0.25  0.25  0.00

       0.25  0.00  0.25

       0.00  0.25  0.25

       0.50  0.00  0.00

       0.00  0.50  0.00

       0.00  0.00  0.50

       0.50  0.50  0.00

       0.50  0.00  0.50

       0.00  0.50  0.50

       0.75  0.25  0.00

       0.25  0.75  0.00

       0.00  0.75  0.25

       0.00  0.25  0.75

       0.75  0.00  0.25

       0.25  0.00  0.75

       0.75  0.75  0.00

       0.75  0.00  0.75

       0.00  0.75  0.75

       0.25  0.25  0.50

       0.25  0.50  0.25

       0.50  0.25  0.25

       0.75  0.75  0.50

       0.75  0.50  0.75

       0.50  0.75  0.75

       0.25  0.50  0.75

       0.25  0.75  0.50

       0.50  0.25  0.75

       0.50  0.75  0.25

       0.75  0.25  0.50

       0.75  0.50  0.25

#Definition of the planewave basis set

ecut  10.0         # Maximal kinetic energy cut-off, in Hartree

pawecutdg 20.0

nband 64

#Exchange-correlation functional

#ixc 11             # GGA

#Definition of the k-point grid

kptopt 1          # Option for the automatic generation of k points

ngkpt  8 8 8

#istwfk 256*1

#nshiftk 1

#shiftk 0.0 0.0 0.0 

#Definition of the SCF procedure

nstep 10           # Maximal number of SCF cycles

#tolvrs 1.0e-15     # Stops SCF cycle when potential residue lower than 

                    # specified number. Is better than toldfe.

toldfe 1.0d-2       # Will stop when, twice in a row, the difference 

                    # between two consecutive evaluations of total energy 

                    # differ by less than toldfe (in Hartree)

# Symmetries of the lattice:

nsym 1

chkprim 0

#*****************************************************************************

____________________________________

Sam M. Vinko

Atomic & Laser Physics

Clarendon Laboratory - University of Oxford

Parks Road, Oxford OX1 3PU, UK

Tel: +44 (0) 1865 272 301

E-mail: sam.vinko@physics.ox.ac.uk