Memory requrements
Posted: Fri Oct 21, 2011 2:53 pm
Hello, I'm trying to compute Si nanoparticle in vasp near the memory limit on our cluster ( 24GB )
One of the jobs fall because of insufficient memory.
There is 108 H-atoms, 120 O-atoms, and 232 Si-atoms with soft pseudopotential ENMAX =269 eV
size of unit cel is 30x30x30 Angstroem ... this leads to
For storing wavefunctions 4826.10 MBYTES are necessary
gives a total of 531441 points
.... this is very much, however if I consider number of planewaves, I was able to run even bigger job: 67H 1O 327Si with normal oxygen pseudopotential ENMAX =395 eV
26x23x36 Angstroem
For storing wavefunctions 5773.98 MBYTES are necessary
gives a total of 885573 points
... this was working, so I don't see reason why the "smaller" job is not
If I look at the allocation output I don't know where all my 24 GB of memory has gone?
- Is there something other than wavefunction which is the memory bottleneck ?
- Can I switch something off in order to save memory ?
this is how the end of OUTCAR looks like
--------------------------------------------------------------------------------------------------------
k-point 1 : 0.00000.00000.0000 plane waves: 271985
maximum and minimum number of plane-waves per node : 34013 33985
maximum number of plane-waves: 271985
maximal index in each direction:
IXMAX= 40 IYMAX= 40 IZMAX= 40
IXMIN=-40 IYMIN=-40 IZMIN=-40
WARNING: wrap around error must be expected set NGX to 162
WARNING: wrap around error must be expected set NGY to 162
WARNING: wrap around error must be expected set NGZ to 162
real space projection operators:
total allocation : 46814.30 KBytes
max/ min on nodes : 5874.90 5837.05
parallel 3dFFT wavefunction:
minimum data exchange during FFTs selected (reduces bandwidth)
parallel 3dFFT charge:
minimum data exchange during FFTs selected (reduces bandwidth)
For storing wavefunctions 4826.10 MBYTES are necessary
For predicting wavefunctions 9814.43 MBYTES are necessary
Broyden mixing: mesh for mixing (old mesh)
NGX = 81 NGY = 81 NGZ = 81
(NGX =240 NGY =240 NGZ =240)
gives a total of 531441 points
initial charge density was supplied:
charge density of overlapping atoms calculated
number of electron 1756.0000000 magnetization
keeping initial charge density in first step
--------------------------------------------------------------------------------------------------------
Maximum index for non-local projection operator 330
Maximum index for augmentation-charges 664 (set IRDMAX)
--------------------------------------------------------------------------------------------------------
<span class='smallblacktext'>[ Edited ]</span>
One of the jobs fall because of insufficient memory.
There is 108 H-atoms, 120 O-atoms, and 232 Si-atoms with soft pseudopotential ENMAX =269 eV
size of unit cel is 30x30x30 Angstroem ... this leads to
For storing wavefunctions 4826.10 MBYTES are necessary
gives a total of 531441 points
.... this is very much, however if I consider number of planewaves, I was able to run even bigger job: 67H 1O 327Si with normal oxygen pseudopotential ENMAX =395 eV
26x23x36 Angstroem
For storing wavefunctions 5773.98 MBYTES are necessary
gives a total of 885573 points
... this was working, so I don't see reason why the "smaller" job is not
If I look at the allocation output I don't know where all my 24 GB of memory has gone?
- Is there something other than wavefunction which is the memory bottleneck ?
- Can I switch something off in order to save memory ?
this is how the end of OUTCAR looks like
--------------------------------------------------------------------------------------------------------
k-point 1 : 0.00000.00000.0000 plane waves: 271985
maximum and minimum number of plane-waves per node : 34013 33985
maximum number of plane-waves: 271985
maximal index in each direction:
IXMAX= 40 IYMAX= 40 IZMAX= 40
IXMIN=-40 IYMIN=-40 IZMIN=-40
WARNING: wrap around error must be expected set NGX to 162
WARNING: wrap around error must be expected set NGY to 162
WARNING: wrap around error must be expected set NGZ to 162
real space projection operators:
total allocation : 46814.30 KBytes
max/ min on nodes : 5874.90 5837.05
parallel 3dFFT wavefunction:
minimum data exchange during FFTs selected (reduces bandwidth)
parallel 3dFFT charge:
minimum data exchange during FFTs selected (reduces bandwidth)
For storing wavefunctions 4826.10 MBYTES are necessary
For predicting wavefunctions 9814.43 MBYTES are necessary
Broyden mixing: mesh for mixing (old mesh)
NGX = 81 NGY = 81 NGZ = 81
(NGX =240 NGY =240 NGZ =240)
gives a total of 531441 points
initial charge density was supplied:
charge density of overlapping atoms calculated
number of electron 1756.0000000 magnetization
keeping initial charge density in first step
--------------------------------------------------------------------------------------------------------
Maximum index for non-local projection operator 330
Maximum index for augmentation-charges 664 (set IRDMAX)
--------------------------------------------------------------------------------------------------------
<span class='smallblacktext'>[ Edited ]</span>