Depending on the number of water molecules, 2 ns may not be long enough; a rule of thumb attributed to M. Klein suggests around 500 molecules may be needed for stable MD simulations.
I've recently used the following for 5 ns with a TIP3P water box of 1340 waters and a ca. 34 A edge (set FFX = 32) and get values around 100, which is what others have reported for this water model.
* PME dynamics with 34 A water box
*
stream rtfprm.str ! RTF, PARAM
stream psfcrd.str ! PSF, COOR
stream cryst.str ! CUBIC LATTICE
shake bonh param
open unit 31 write file name dyn.trj
open unit 40 read card name dyn.rea
open unit 41 write card name dyn.res
! NVT ENSEMBLE WITH PRESSURE REPORTING (PMASS 0)
dyna cpt leap restart echeck 250. nstep 500000 nprint 500 -
iprfrq 5000 atom cdie cutnb 16.0 ctofnb 12.0 -
wmin 1.5 eps 1.0 inbfrq -1 cutim 16.0 imgfrq -1 -
ewald pme order 6 spline kappa 0.32 fftx @FFX ffty @FFX fftz @FFX -
pcons pmass 0.0 pgamma 0.0 pref 1.0 -
hoover reft 293 first 293 finalt 293 -
ihtfrq 0 ieqfrq 0 ntrfrq 500 ichecw 0 iasors 1 -
nsavc 1000 iuncrd 31 iunrea 40 iunwri 41
stop
WARNING: the TCONS keyword was originally included by mistake in the indicated line above; DO NOT use that keyword in conjunction with HOOVER 
