GPU acceleration in electron-phonon renormalization calculation of VASP 6.5.1

[jiong_yang1]

Dear all,
I hope this message finds you well. I am currently using the electron-phonon renormalization calculation of VASP 6.5.1.During band structure renormalization calculations, VASP performs KPOINTS_OPT, with outputs such as:

Code: Select all

elph_comm: one band on    1 cores,    1 groups
k-point batch [   110 /    221 ] workload [   220 /    441] on node [     2 /      2]
Start KPOINTS_OPT (optional k-point list driver)
k-point batch [   111 /    221 ] workload [   221 /    441] on node [     1 /      2]

However, based on my observations, KPOINTS_OPT does not seem to be accelerated by GPUs, as there is no GPU utilization during this step. Could you confirm if this is the expected behavior?

Thank you sincerely for taking the time to address these questions. I look forward to your detailed responses.
Best regards

[christopher_sheldon1]

Dear Jiong,

Thank you for your question. You are correct that there is no GPU utilization in electron-phonon band structure renormalization calculations, so there is also no acceleration by using machines with GPUs.

Does this answer your question?

Best wishes,

Chris

[jiong_yang1]

Thanks, that helps me a lot. By the way, we also find 2 more questions in electron-phonon renormalization calculation of VASP 6.5.1.
[1] During the calculation of graphene renormalization, the phelel_params.hdf5 file generated in the first step comes from the calculation of a 50-atom supercell with a k-mesh of 5*5*1, and its size is approximately 140 MB. The unit cell POSCAR in the second step contains two atoms. When the unit cell k-mesh is 15*15*1, VASP will report the following error:

Code: Select all

 ========== interatomic force constants  ============
 treatment of LR: F
 ====================================================
 =========== electron-phonon potential  =============
 treatment of LR: F
 shmem: F
 spin-unpolarized (ispin=1, nrspinors=1, ncdij=1)
 ====================================================
 ============== electron-phonon wfs =================
 fft_mesh:             20          20         120
 spin-unpolarized (ispin=1, nrspinors=1, ncdij=1)
 nbands_k':          2691
 nbands_k:           2691
 elph_nbands:        2691
 ====================================================
 -----------------------------------------------------------------------------
|                                                                             |
|     EEEEEEE  RRRRRR   RRRRRR   OOOOOOO  RRRRRR      ###     ###     ###     |
|     E        R     R  R     R  O     O  R     R     ###     ###     ###     |
|     E        R     R  R     R  O     O  R     R     ###     ###     ###     |
|     EEEEE    RRRRRR   RRRRRR   O     O  RRRRRR       #       #       #      |
|     E        R   R    R   R    O     O  R   R                               |
|     E        R    R   R    R   O     O  R    R      ###     ###     ###     |
|     EEEEEEE  R     R  R     R  OOOOOOO  R     R     ###     ###     ###     |
|                                                                             |
|     wf cache is full                                                        |
|                                                                             |
|       ---->  I REFUSE TO CONTINUE WITH THIS SICK JOB ... BYE!!! <----       |
|                                                                             |
 -----------------------------------------------------------------------------

This calculation was performed on a 24-core CPU with 512 GB of memory. Could the reason for the error reported here be insufficient memory, or is it another hardware issue? Below is my INCAR:

Code: Select all

PREC = Accurate   
ENCUT = 600     
EDIFF = 1E-8    
LREAL = .FALSE. 
ISMEAR = 0   
SIGMA = 0.01 eV    
LWAVE = F
NCORE = 2

ELPH_RUN = .TRUE.    
ELPH_DRIVER = EL   
ELPH_NBANDS = -2      
KPOINTS_OPT_MODE = 1   
ELPH_IGNORE_IMAG_PHONONS = T

ELPH_SELFEN_GAPS = F 
ELPH_SELFEN_DELTA = 0.01 
ELPH_SELFEN_FAN = .TRUE.  
ELPH_SELFEN_DW = .TRUE.    
ELPH_SELFEN_ENERGY_WINDOW = 1.500 5.500
ELPH_SELFEN_TEMPS = 0 100 200 300 400 500 

[2] Furthermore, if it is indeed due to hardware issues that dense k-mesh electron-phonon renormalization self-energy calculations cannot be performed, can Fourier interpolation calculations be carried out instead?

I am eagerly looking forward to your reply!
Best wishes

[jiong_yang1]

Oops, it seems that my problem has been solved, and no further answers are needed for subsequent questions for the time being. Thank you very much indeed!

[christopher_sheldon1]

Glad that you solved your problems. I'll lock the topic on Friday afternoon, unless you have any further questions before then.

Best wishes,

Chris