The short answer is that you want to use the latter (sigma->0).
There's a bit of an explanation at http://cms.mpi.univie.ac.at/vasp/vasp/ISMEAR_SIGMA_FERWE_FERDO_SMEARINGS_tag.html, but the basic idea is that, if you're using Gaussian smearing, the total energy will only be correct in the limit sigma->0. The latter value is then the extrapolated total energy for zero smearing. For things like the tetrahedron method (ISMEAR=-5), both values should be identical.
The short answer is that you want to use the latter (sigma->0).
There's a bit of an explanation at http://cms.mpi.univie.ac.at/vasp/vasp/ISMEAR_SIGMA_FERWE_FERDO_SMEARINGS_tag.html, but the basic idea is that, if you're using Gaussian smearing, the total energy will only be correct in the limit sigma->0. The latter value is then the extrapolated total energy for zero smearing. For things like the tetrahedron method (ISMEAR=-5), both values should be identical.
In practice it probably doesn't matter which one you choose, as long as you are consistent. The absolute value of a DFT total energy is not very meaningful, usually the total energy difference between various configurations and/or conditions are considered.
There is a continuous discrepancy between the vasp energies. As far as i know, the vasp calculations are done at an elevated temperature (not at exactly zero) for the fastness of the calculations. The elevation in the TEMP is maintained by "SIGMA" values. If u choose larger sigma value, the inclusion of temperature in the calculation will be more, and a very smaller sigma value will give the energies similar to the energies which are performed at zero temperature. But u need the energies exactly at Zero temperature and hence the Sigma is extrapolated to zero, which means that energy is equal to the energy at zero temperature. Generally, the molecules at lower temp will be more stable than that of higher temp. Hence ur ( sigma--> 0) value will be always smaller than ur original value