it achieves it in a relative manner. DInSAR is not an absolute measurement principle. So what you actually measure is the relative change in distance from time 1 to time 2. You assuming your DEM therefore as "correct", or in a more technical way, you use it to substract the topographical component of your interferometric phase. In addition, 2-pass DInSAR does not achieve necessarily sub-centimeter accuracy. Atmospheric conditions might have strong effects (e.g. 14cm for 20% change in relative humidity). If you want to achieve sub-centimetres accuracy, go for multi-temporal approaches like SBAS and/or PS. Those techniques provide you also an DEM error estimation as an additional output. There are some freely available open-source software packages out for research purposes. StaMPS or GiANT for example.