Handling memory on GW

[sponza]

Hi to everybody.
I am facing a tough GW calculation (big cell, many many bands, ~100 electrons). I have a WAVECAR of 7.1 Gb and a WAVEDER of 316 Mb. When I run the ALGO=GW0 step, I get stuck at the calculation of Chi0. No understandable error is displayed but I have strong indications that is a memory problem.

So my questions are

1) Let' imagine I don't care about calculation time, how can I reduce the memory consumption without lowering the parameters? (e.g. I tried LSPECTRAL = FALSE, but I haven't succeded)

2) How can estimate the memory I will need or monitor the memory I'm using? The queue system is managed by PBS.

My INCAR file is the following:
NPAR = 16
ALGO = GW0
LSPECTRAL = True
LOPTICS = True
LRPA = True
PRECFOCK = FAST
PREC = Normal
ENCUT = 600.00 eV
NOMEGA = 100
NBANDS = 3000
ENCUTGW = 560.00 eV
ODDONLYGW = True
CSHIFT = 0.1
LMAXFOCKAE = 4
NBANDSGW = 96
NKRED = 2
MAXMEM = 4000
ISMEAR = 0 ; SIGMA = 0.05


A typical OUTCAR file ends at this point:
total amount of memory used by VASP on root node 8647300. kBytes
========================================================================

base : 30000. kBytes
nonl-proj : 193511. kBytes
fftplans : 1377. kBytes
grid : 4409. kBytes
one-center: 406. kBytes
HF : 423. kBytes
nonlr-proj: 1695. kBytes
wavefun : 1635581. kBytes
2eintegral: 23101. kBytes
response : 6756797. kBytes



--------------------------------------------------------------------------------------------------------


NQ= 1 0.0000 0.0000 0.0000,



Thanks a lot for any help/suggestion/compassion.

Lorenzo

[jxzhu]

Hello,

In what units the MAXMEM=4000 is in, Mb?

Thanks,

Jianxin