Deallocate error 492: Array is not allocated

[tangxin]

Dear all:
I recently want to optimize geometric structure of interstitial atom in ZnO. My INCAR file is
SYSTEM = ZnO
ISTART=0
PREC=M
GGA=91
ENCUT=400
NGX=66;NGY=66;NGZ=72
LREAL= Auto
NELM = 200
NELMIN = 3
EDIFF = 1E-5
EDIFFG = -0.01
IALGO = 48
LWAVE=F

NSW=400
ISMEAR = -5
IBRION = 2
ISIF = 3

LVTOT=.TRUE.
ISPIN=1
LDAU=.TRUE.
LDAUTYPE=4
LDAUL=2 -1
LDAUU=7.0 0.0
LDAUJ=0.0 0.0
LDAUPRINT=1

LPLANE=.TRUE.
NPAR=36
LSCALU=.FALSE.
NSIM=4
POTIM=0.3

I used 36 cpus to calculate. My machine is Linux cluster and Vasp is compiled by IFC 7.0. There was no error at the beginning but the calcultion stopped in the midway of the second SCF. At one time, "Deallocate error 492: Array is not allocated" appeared.
Based on the handbook of IFC and searching result on the internet, I found that "deallocate 492 error occurs when total size of allocated arrays too large".
I also set NPAR=4 to calculate again(using 36 cpus) but the error cannot be avoided.

So, I want to know how to avoid the error. Does the error mean the size of memory is not enough?

Thanks!!