It's important to understand the columnar layout and internal storage of the import mode datasets. Power BI creates individual segments of approximately one million rows and stores separate memory structures for column data, the dictionary of unique values for columns, relationships, and hierarchies.
In the following diagram, three segments are used to store a fact table of 2.8 million rows:
Columnar storage of import mode datasets
Since only the columns required for a query are scanned during query execution, a relatively expensive column in terms of memory consumption (due to many unique values), such as Order #, can ...
