Huge memory footprint with nonlocal pseudopotentials
Posted: Fri Jul 11, 2014 6:05 am
I am trying to run an aluminum nanocluster containing 666 atoms(5x5x5 fcc cells).
I am enclosing the nanocluster in a box of size 40 a.u where the cluster size itself is 37.25 a.u. I am trying to find the ground state energy using a non-local norm-conserving TM pseudopotential. I am facing serious memory issues with ABINIT code. Even when I ran ABINIT on 64 procs with each proc having 64 GB of memory, my computing nodes ran out of memory. Infact I was also using nloalg = -4 and paral_kgb = 1. It did not even complete one SCF iteration. I am wondering if I am missing anything here. This is happening especially when I use nonlocal pseudopotentials. If I suppress the nonlocal part, then I do not face memory issues...
I am enclosing the nanocluster in a box of size 40 a.u where the cluster size itself is 37.25 a.u. I am trying to find the ground state energy using a non-local norm-conserving TM pseudopotential. I am facing serious memory issues with ABINIT code. Even when I ran ABINIT on 64 procs with each proc having 64 GB of memory, my computing nodes ran out of memory. Infact I was also using nloalg = -4 and paral_kgb = 1. It did not even complete one SCF iteration. I am wondering if I am missing anything here. This is happening especially when I use nonlocal pseudopotentials. If I suppress the nonlocal part, then I do not face memory issues...