Threads issues on Bethe-Salpeter solver-diagonalization

Various technical topics such as parallelism and efficiency, netCDF problems, the Yambo code structure itself, are posted here.

Moderators: Davide Sangalli, andrea.ferretti, myrta gruning, andrea marini, Daniele Varsano, Conor Hogan, Nicola Spallanzani

Post Reply
ljzhou86
Posts: 85
Joined: Fri May 03, 2013 10:20 am

Threads issues on Bethe-Salpeter solver-diagonalization

Post by ljzhou86 » Sun Jan 29, 2023 4:03 am

Dear developers

I utilized the bulk-hBN examples to test the latest version 5.1.1 on the BSE calculation by using Bethe-Salpeter solver-diagonalization.

I encountered a strange issue: that is, the calculation successfully passed the BSK diagonalization when I set a threads value of 2 on DIP_Threads, X_Threads and K_Threads; however, it got stuck without any error information when I used a threads value of 4. What is the reason? Are there any rules on the settings for a good pass?

Please see the attached input, report and LOG files.

Thanks a lot.
You do not have the required permissions to view the files attached to this post.
Dr. Zhou Liu-Jiang
Fujian Institute of Research on the Structure of Matter
Chinese Academy of Sciences
Fuzhou, Fujian, 350002

User avatar
Daniele Varsano
Posts: 3773
Joined: Tue Mar 17, 2009 2:23 pm
Contact:

Re: Threads issues on Bethe-Salpeter solver-diagonalization

Post by Daniele Varsano » Mon Jan 30, 2023 9:12 am

Dear Zhou,

I'm not sure the problem here is the number of threads. The main difference between the two calculations is that in the case threads=4 parallel linear algebra is activated.
I do not think that this is due to the number of threads, but most probably to the number of used CPU (12 for threads=2 and 24 for threads=4).
You can try to use the same number of CPU for the two cases. Or set the serial linear algebra also for threads=4 by setting:

Code: Select all

BS_nCPU_LinAlg_DIAGO=1
If you want to use many threads, maybe you want also to set up the number of MPI consistently in order to avoid a too large hyper-threading.

Best,
Daniele
Dr. Daniele Varsano
S3-CNR Institute of Nanoscience and MaX Center, Italy
MaX - Materials design at the Exascale
http://www.nano.cnr.it
http://www.max-centre.eu/

Post Reply