The numeric types float, double precision, and real store numeric data that can tolerate rounding during arithmetic operations.
Approximate numeric datatypes store, as binary tractions, slightly inaccurate representations of real numbers, stored as binary fractions. Anytime an approximate numeric value is shown, printed, transferred between hosts, or used in calculations, the numbers lose precision. isql displays only six significant digits after the decimal point, and rounds the remainder. See System and User-Defined Datatypes, in the Reference Manual: Building Blocks.
Use the approximate numeric types for data that covers a wide range of values. They support all aggregate functions and all arithmetic operations.
The real and double precision types are built on types supplied by the operating system. The float type accepts an optional precision in parentheses. float columns with a precision of 1 – 15 are stored as real; those with higher precision are stored as double precision. The range and storage precision for all three types is machine-dependent.