Although SVD and SDD work in different ways, if a dataset contains a genuine clustering, it should be visible to both algorithms. SVD and SDD are quite complementary. SVD is able to make the most important structure visible in the early dimensions, but it is hard to exploit this directly because there are multiple ways to construct and label clusters from it. SDD, on the other hand, tends to produce more, smaller clusters than SVD (because they are really biclusters) but provides an automatic labelling of objects with the cluster they belong to, using some subset of the columns of X.