Relaxing problem for Pd cluster (2atoms)

[s9811016]

I have specify NSW = 120 steps, EDIFF = 1E-05 (tried -03 to 07) and EDIFFG = -0.01 ( tried to -.001). but after 12 step the calculation is stopped, Anybody knows where is the problem?
Here is my INCAR
Pd_cluster:
Startparameter for this run:
NWRITE = 2
ISTART = 0
ICHARG = 2
PREC = medium ! standard precision

Electronic Relaxation:
ENCUT = 400 ! cutoff should be set manually
EDIFF = 1E-05
LWAVE = .FALSE.
LCHARGE = .TRUE.
LREAL = F ! projection done in real space
Ionic Relaxation:
IBRION = 2
ISIF = 3
NSW = 120
EDIFFG = -0.01
ISYM = 2
POTIM = 0.4

Parallelization flags
NPAR = 1
LPLANE = .TRUE.

DOS Related values:
ISMEAR = 0 ; SIGMA = 0.1 ! O for GS,-5 for Tetra +Blochl& broadening in eV



11 F= -.10848101E+02 E0= -.10848101E+02 d E =-.730647E+01
curvature: -0.16 expect dE=-0.409E-05 dE for cont linesearch -0.409E-05
ZBRENT: interpolating
opt : 12.8946 next Energy= -10.848105 (dE=-0.731E+01)
bond charge predicted

12 F= -.10848106E+02 E0= -.10848106E+02 d E =-.730648E+01
curvature: -0.14 expect dE=-0.865E-06 dE for cont linesearch -0.865E-06
trial: gam= 0.00000 g(F)= 0.153E-62 g(S)= 0.617E-05 ort =-0.991E-03 (trialstep = 0.338E+01)
search vector abs. value= 0.617E-05
reached required accuracy - stopping structural energy minimisation

[admin]

1) NEVER use ISIF 3 if you do a small cluster in a large unit cell.
2) check for the remaining forces in OUTCAR, if the convergence limit is reached, vasp stops.