In-memory, the Future - Getting Started with SQL Server 2012 Cube Development - page 261

By storing data in each column separately, a higher compression rate can be
achieved compared to applying compression on the row level. This is illustrated
in the preceding screenshot. The 15 rows have the same ProductKey , the same
SalesTerritoryKey , and the same SalesAmount , and hence, this value only
needs to be stored once. There are also duplicates among the DateKey columns
and the duplicate rows do not need to be stored.
The effect of this compression will both be lower space consumption as well as high-
er performance, because the server needs to scan less data to calculate the queries.
This means that you do not need to work with aggregations to get better perform-
ance from your model.