In addition to using Packmol to construct the box with randomly positioned H2O molecules, a simple Force-Field based geometry optimization (minimization of both forces on atoms and stress on the supercell) might give a good starting point for DFT.
Flow:
1. use Packmol to construct the box with randomly positioned H2O molecules (perhaps aim for experimental density of molecules).
2. change cell type from UnitCell to SimpleCubic.
3. use ForceField calculator and OptimizeGeometry to relax forces and stress (take care to select a force field that actually works for water).
Example script attached. Runs in 2 mins on a laptop. Resulting water structure looks pretty OK to me, especially with the expected directional non-covalent bonding between H and O on different molecules.
If you believe finite-temperate molecular dynamics is needed to get better or more realistic water structure, that is also an option, but less simple than flow given above.