Error of out-of-memory

Zafer Kandemir · Post by **Zafer Kandemir** » Sat Apr 24, 2021 2:30 pm

Dear Developers,

I'm trying to perform the GW (real-axis) calculations by using Yambo (Version 5.0.0). I am working on a server with the information below.
-------------------------------------
Server Information:
---------------------------------
Pool Memory Cpus
---------------------------------
sardalya 256000Mb 28
barbun 384000Mb 40
-------------------------------------
I started a GW calculation by using the 28 cores. But my calculation stopped with an insufficient memory problem.
How can I set the SE_ROLEs part?
Do you have any suggestions to overcome such a problem? Thanks.

Best regards.
Zafer

I am attaching all the logic, report, slurm-jobid.err and slurm-jobid.out files for my calculation.
I also attached input file for my calculation.

Post by **claudio** » Sat Apr 24, 2021 7:57 pm

Dear Zafer

please add the Yambo input file used for the calculation.
A possible strategy to decrease the amount of memery in the GW is to put all parallelization
on the "G bands" , and do not parallelize on the Quasiparticle.

best
Claudio

Zafer Kandemir · Post by **Zafer Kandemir** » Tue Apr 27, 2021 12:35 pm

Dear Claudio,

Thank you for your advice. I added the Yambo input file.

Post by **Daniele Varsano** » Tue Apr 27, 2021 1:02 pm

Dear Zafer,
the input file is not attached.
In order to upload it you need to rename it adding a supported suffix (e.g. .txt)
Best,
Daniele

Zafer Kandemir · Post by **Zafer Kandemir** » Tue Apr 27, 2021 7:44 pm

Dear Daniele,

I edited the first post and attached the input file. I added it here again. Thanks.

Zafer

Post by **Daniele Varsano** » Wed Apr 28, 2021 12:41 pm

Dear Zafer,
real-axis calculations are extremely heavy, also because it is not guaranteed that 100 frequencies are enough to get converged results.
Usually plasmon-pole approximation works very well.
Anyway, if you need to go through a real axis calculation, as Claudio said, my suggestion is to parallelise over bands:
SE_CPU="1.1.28"
SE_ROLEs="q.qp.b"

and next split your calculation in several runs e.g:
first run:

Code: Select all

%QPkrange                        # [GW] QP generalized Kpoint/Band indices
1|10|27|30|
%

second run

Code: Select all

%QPkrange                        # [GW] QP generalized Kpoint/Band indices
11|20|27|30|
%

etc.

renaming te ndb.QP database of each run, or assigning a different label, at the end you can build a single database using ypp to merge them.
Best,
Daniele

Zafer Kandemir · Post by **Zafer Kandemir** » Wed Apr 28, 2021 7:01 pm

Dear Daniele,

Thank you very much for your advice. I read the topics "Merging databases" and "Splitting a GW run then merge the database" about ypp. I will try it.

Best regards.
Zafer

Yambo Community Forum

Error of out-of-memory

Error of out-of-memory

Re: Error of out-of-memory

Re: Error of out-of-memory

Re: Error of out-of-memory

Re: Error of out-of-memory

Re: Error of out-of-memory

Re: Error of out-of-memory