The tutorial is still available physically (
here) but as the title etc indicates, it applies mostly to the older versions of the software, and to some extent to the specific FeMgO system which was hard to converge before we introduced the double contour integration and a more correct treatment of the spill-in terms.
In ATK 12.2 and later, convergence problems are almost invariably related to either too few k-points or a too short electrode (or, perhaps, too short screening region), unless of course the geometry is unusual and complicated.
At high bias you need a longer screening region. A rather special trick which can also help at high bias is to use a smaller value for the "real_axis_point_density".