Dask is not necessarily a good way to scale up your model training—most models require interaction, and therefore should stay within one machine. At the same time, most sklearn models can work on multiple CPUs on their own, and so Dask isn't required.

With that being said, there are plenty of cases when using Dask could be beneficial. For that, there is an additional layer over Dask—Dask-ML. Dask-ML helps connect Dask to sklearn and other ML libraries (for example, XGBoost and TensorFlow), thereby allowing you to run some parallelizable models (linear models, for example, or some clustering algorithms), execute hyperparameter searches with different hyperparameters being executed on different servers, or connect distributed datasets to large modules, such as XGBoost.