Intervals in interval arithmetic can rapidly go to infinity even when the problem is relatively well behaved (in matrix inversion, for example) and standard floating point arithmetic produces a useful answer.