My Community

For this system
http://www.fkp.uni-hannover.de/~meyer/N ... system.tgz
and some similar that I have tried VASP crashes with a segmentation fault
after the first ionic step on several architectures (Athlon, Opteron, Power4)
when NSW is set 'too high'.
The critical value for this system is different from system to system
(Athlon: >=415, Opteron: >=464, Power4: >500 and <1000).
In the supplied INCAR file NSW is set 1000 which should trigger this behaviour
anyway. A (converged) WAVECAR file is supplied to fasten up experimenting
with different values for NSW.
Of course, 1000 are rather many ionic steps. However, more around
500 steps may be necessary for systems like slabs for simulation of the
adsorption of complex molecules (~ 50 non fixed atoms...).

sorry, I could not reproduce your crash with NSW=1000 on any of the machines I tried (PentiumIV, Opteron). I rather think that the crashes are not related to NSW itself but to numerical instabilities to find the next trial step of a fully converged system.
I can therefore only give you the advise to try one of the following:
-- concerning NSW: split your job into a sequence of calculations with a 'safe' choice of NSW (and continue from WAVECAR and CONTCAR (copied to POSCAR) of the last step of the previous run
-- try a different ionic relaxation algorithm or
-- set NFREE to the number of steps which you want to keep in the 'relaxation history'