Unfortunately, I don't have experience with VASP.

I hope to be more useful for you in the future.

Regards,

                        PAblo

Dear Garam,

It is probably possible to construct a reasonable model of the time per SCF cycle, but estimating how many SCF cycles you will need would be difficult -- and that's without considering how many geometry optimisation steps you might need if you're moving the atoms or lattice vectors.

Why do you want to predict the run-time? I am not familiar with VASP in particular but I do develop a plane-wave DFT program so do know a bit about the issues involved. The main reason we wanted such a model in our code (CASTEP) is to predict whether the calculation would finish before the walltime limit in a compute queue so we could write a checkpoint file, but we achieved this by timing the iterations directly. A good model of run-time would be very useful for optimising and automating the choice of parallelisation, so it is something I'm looking into at the moment.

All the best,

Phil Hasnip

(CASTEP developer)

I use public server in my univ. that's why I want to know the finish time. 

Thank you and hope you make your code like that!! :)

Sincerely,

Garam Choi