Yambo Community Forum

Dear developers,

I have run a same GW case on a CPU HPC and GPU machine and find that the time consuming is completely different in each part. Here I list the configuration of each machine and time consuming in each part.

• CPU HPC: 32MPI+8OMP (256cores total)
  • Dipoles: 19s
  • Dynamic Dielectric Matrix (PPA): 01h-04m
  • Local Exchange-Correlation + Non-Local Fock: 03h-20m
  • Dyson equation: 02d-09h-00m

• GPU HPC: 8CPU cores+8GPU (Tesla A100)
  • Dipoles: 38s
  • Dynamic Dielectric Matrix (PPA): 19h-21m
  • Local Exchange-Correlation + Non-Local Fock: 13m-42s
  • Dyson equation: 01d-19h-34m

Obviously, the dielectric matrix perform totally differently on CPU and GPU machine. Is there any good way to improve the GPU performance on this part? Are there any special input parameters to set in the input files for a GPU machine? Thank you!

Best,
Jason

Dear Jason,
could you send us the log files of both the calculations?

Best,
Nicola

Dear Nicola,

Thanks for your reply. Here attached the two log file. Due to the file size limit, I delete some useless information in the log file. Hope it works.

Best,
Jason