incorrect number of electrons in outcar with intel compiler

[cpp6f]

I'm seeing a problem on at least three different computer clusters where the number of electrons calculated in each scf step is wrong. For example, the line in the outcar right before the calculation begins is

number of electron 368.0000000 magnetization

but the first scf step says

number of electron 419.7855588 magnetization
augmentation part 419.7855588 magnetization

needless to say, the calculation doesn't converge.

This only happens when using the intel compiler. When I switch to pathscale or pgi, it goes away. The math library doesn't seem to make a difference either since I get the error with the intel compiler whether I use mkl or gotoblas. Doesn't seem to depend on the mpi library either. It seems to depend on the machine architecture also since the intel compiled vasp works on two other clusters. I'm using 4.6.34 version of vasp

Has anyone else experienced this error before? The clusters I am using are chinook at pnl, and lonestar and ranger at tacc.
<span class='smallblacktext'>[ Edited ]</span>

[admin]

which intel compiler release do you use? usually these types of errors do NOT show up with intel-compiled vasp, if vasp is installed correctly (ie all loaded libraries are implemented correctly,...)