Hi, I'm starting a Data Mart from scratch for a HR database. In a traditional DW model, I would have organized data in Facts and Dimensions. However, I don't know if Power BI would benefit from that or it would be better to directly organize data in flattened tables, I mean, denormalizing all data and putting dimensions valued directly in the fact tables. I would keep only "special" dimensions separated like Time and Geographical data.
There are several blog posts out there indicating that Power BI is faster with dimensions versus everything being flat. Remember, under the hood, Power BI is a tabular cube so normal design rules for tabular cubes still apply.
Proud to be a Datanaut!