Strange parallel behavior for GW calculation

[Sheleon]

Hi everyone,
When i use yambo calculate the COHSEX correction using 8 cores, it goes well until the COHSEX calculation begin to start. The strange parallel behavior is that when come into COHSEX calculation, 7 cores are running aproperiately, name using 100%, but the last core is only using 20%-40% result in slowing down the whole calculation.
Here is my input file and log file.

Code: Select all

#             GPL Version 3.4.1 Revision 3187               
#                http://www.yambo-code.org                  
#
em1s                         # [R Xs] Static Inverse Dielectric Matrix
gw0                          # [R GW] GoWo Quasiparticle energy levels
cohsex                       # [R Xp] COlumb Hole Screened EXchange
HF_and_locXC                 # [R XX] Hartree-Fock Self-energy and Vxc
EXXRLvcs= 15        Ry    # [XX] Exchange RL components
Chimod= "Hartree"            # [X] IP/Hartree/ALDA/LRC/BSfxc
% BndsRnXs
   1 | 36 |                 # [Xs] Polarization function bands
%
NGsBlkXs= 5            Ry    # [Xs] Response block size
% LongDrXs
 1.000000 | 0.000000 | 0.000000 |        # [Xs] [cc] Electric Field
%
#% GbndRnge
#  1 | 200 |                 # [GW] G[W] bands range
#%
%QPkrange                    # [GW] QP generalized Kpoint/Band indices
  1| 68|  27|37|
%
%QPerange                    # [GW] QP generalized Kpoint/Energy indices
  1| 68| 0.0|-1.0|
%

Thanks in advance

Sheleon

[Daniele Varsano]

Dear Sheleon,
there could be many reason for that, and it is not easy now to spot the problem. I can't see any strange stuff in your log file: from your log it looks that the cohsex calculation was not started. The only think I can see is that it is a quite memory consuming calculation (~7.5Gb) until it stopped. So it could also be memory reason and 1 cpu get stacked.
In any case, we are totally changing the parallel procedure for the next release of yambo, that we hope to release soon. For this reason we will not work on parallelization issues of the actual release.
Best,
Daniele

[Sheleon]

Dear Danilel,
Thank you.

Sheleon

[Daniele Varsano]

Dear Sheleon,
anyway my impression is that you are exceeding the available memory per cpu. Of course it is just an impression that would need a check, you can try to low some of your convergence parameter.
Best,
Daniele

[Sheleon]

Dear Daniele,
I indeed low the exchange and screen wave vectors to check out. However, the strange behavior is still there, but the calculations can goes on with low parameters.

Sheleon