Memory issue with full gw claculation

[Ponnappa]

Dear Yambo Developers and Users,

I am facing memory issues when considering full GW band structure. I have access to 4 nodes with 32 cpu each. I have tried various staterigies to parallelize with c and v but keep getting the same issue(I also face same problem in first step of BSE calculaiton). I am attaching the r_setup

r_setup.txt

, input file

gw.in.txt

and the LOG file

l-jdir_HF_and_locXC_gw0_dyson_rim_cut_em1d_ppa_el_el_corr_CPU_1.txt

. I face the memory issue very often. I would request you to help me in getting insights about best stratergies to parallelize and solve the issue. Thank you in advance.

Regards,
Ponnappa K P

[Daniele Varsano]

Dear Ponappa,

in order to distribute memory efficiently, you can set something like:

Code: Select all

X_and_IO_CPU= "1 1 1 12 10"                 # [PARALLEL] CPUs for each role
X_and_IO_ROLEs= "q,g,k,c,v"               # [PARALLEL] CPUs roles (q,g,k,c,v)

running with 120 code. If this is not enough you need to run with fewer cpus per node, allowing in this way more memory per mpi-task.

Also note that you cannot set in "b" more CPU than the ones indicate in GbndRnge:

Code: Select all

SE_CPU= "1 1 128"                       # [PARALLEL] CPUs for each role
SE_ROLEs= "q,qp,b"     

Best,
Daniele