I am searching of how to set the tags NCORE, KPAR and NBANDS on a specific systems to speed up calculations. Could you please tell me the rules to follow ?


Well This question is very much demanding for all user, How to reduce calculation time without loosing accuracy of calculation. There are several ways suggested by various VASP user but at the end it all depends on the architecture of the machine. Factor that need to be addressed first are:

  1. Number of CPU per node
  2. Number of Node available
  3. whether GPU available
  4. Communication between node
  5. Intel vs open libraries

Hence its bit tricky to answer fully that how much resource I need for optimum result. one can easily benchmark their cluster/machine by simple scf calculation. or just follow https://github.com/JMSkelton/VASP-GPU-Benchmarking. Let me discuss what is commonly observed results, KPAR= number of nodes, AlGO=Fast (RMM-DIS also fails in many complex system) NCORE= determines how many cores work on one orbital (Try 4,8, up to number of core per node to find optimal) Generally NCORE=8 or 16 is enough. NPLANE,NSIM,NBANDS are three other parameter you can play in your system but keep checking on amount of memory calculation is demanding. There are several studies has been done on this

  1. https://www.nsc.liu.se/~pla/blog/archives/
  2. https://www.nvidia.com/en-us/data-center/gpu-accelerated-applications/vasp/
  3. http://www.hector.ac.uk/support/documentation/software/vasp/benchmark/

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.