Author Topic: Calculation using Plane wave method(beta version)  (Read 3792 times)

0 Members and 1 Guest are viewing this topic.

Offline Inaoton

  • QuantumATK Guru
  • ****
  • Posts: 114
  • Country: in
  • Reputation: 0
    • View Profile
Calculation using Plane wave method(beta version)
« on: August 9, 2017, 09:01 »
 
    I start a calculation using plane wave method as in ATK 2017.0 version .  The script was running for a while  but suddenly the GUI interface-VNL crashes completely. Is it related to some sort of software problem or some memory requirement in RAM?

Offline Jess Wellendorff

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 933
  • Country: dk
  • Reputation: 29
    • View Profile
Re: Calculation using Plane wave method(beta version)
« Reply #1 on: August 9, 2017, 09:14 »
We would need to see the script to be sure, but most likely a memory problem.

Offline Inaoton

  • QuantumATK Guru
  • ****
  • Posts: 114
  • Country: in
  • Reputation: 0
    • View Profile
Re: Calculation using Plane wave method(beta version)
« Reply #2 on: August 9, 2017, 11:30 »
The script is attached
Thank you

Offline Jess Wellendorff

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 933
  • Country: dk
  • Reputation: 29
    • View Profile
Re: Calculation using Plane wave method(beta version)
« Reply #3 on: August 9, 2017, 13:56 »
The calculation takes at leas 4.8 GB per process. If you run it in parallel, the RAM usage increases correspondingly. I suggest running your PW calculation on a supercomputing cluster, with maximum MPI parallelization over k-points.

Offline Anders Blom

  • QuantumATK Staff
  • Supreme QuantumATK Wizard
  • *****
  • Posts: 5577
  • Country: dk
  • Reputation: 96
    • View Profile
    • QuantumATK at Synopsys
Re: Calculation using Plane wave method(beta version)
« Reply #4 on: August 15, 2017, 05:40 »
Moreover I suggest you wait a little bit to run serious plane-wave calculations in ATK - the upcoming 2018 release already provides performance and memory improvements which make the code competitive with other codes out there, however the 2017 edition is slow and not feature-complete (no noncollinear spin for instance).