Lammps is simulated with too much memory

[manbujin]

Hello！
I trained the potential function of nine elements, and when I did a lammps simulation of 1,700,000 atoms, it took up almost 1T of memory. I've compressed the dp model, but it still takes up a lot of memory, is there any solution for this?
Thank you very much for your answer.

[njzjz]

You can consider setting the precision to float32 to save memory. xref: deepmodeling/deepmd-kit#3805

[manbujin]

Thank you very much, it effectively reduces the memory footprint.