Teradata recommends that you explicitly specify a PI for most tables without column partitioning and a PA for most column-partitioned tables. However, there are certain types of tables and situations for which a PI or PA is unnecessary and undesirable, due to index processing overhead or an appropriate set of index columns is not known. In these cases, tables can be created as no-primary-index (NoPI) tables.
For example, staging tables, to which bulk data is batch loaded, are typically temporary. After loading, the data is typically transferred to a PI or PA table using INSERT ... SELECT statements.
For an even distribution of NoPI table rows among AMPs, the rows (or blocks of rows) are randomly distributed during a load. This helps distribute rows evenly among the AMPs. For an INSERT ... SELECT statement, rows are locally transferred by default, which may cause skew. You can use a HASH BY clause to increase the chance of an even distribution of rows.