VASP versus ABINIT for parallel calculations [Archive]

olgranpappy

Oct20-09, 03:50 PM

Does anyone here have experience using VASP (especially VASP 5.2) for parallel calculations? I'm currently using ABINIT (abinip, and abinip with the "paral_kgb" variables) and although the price is right (free), it has been giving me some problems lately.

The input files for VASP and ABINIT seem to have a fairly similar structure, so I was thinking about buying VASP. Anyone have any advice in this regard? For example, how does ABINIT compare to VASP as far as large parallel jobs (~100--1000 processors) go?

Thanks,

Adam

Modey3

Oct23-09, 06:23 PM

olgranpappy,

Did I read you right? 100-1000 processors? That's a lot of processors. Sure each processor can diagnolize a portion of the Hamiltonian matrix, but each processor needs to communicate with each other, which will bottleneck your calculation since there are so many processors. The most processors I've ever used was 8 processors for about 125-atoms.