SQL> create table fnd (num number,num_float binary_float,num_double binary_float);
 
Table created

SQL> insert /*+append*/ into fnd select rownum,rownum,rownum from dba_objects;
 
50070 rows inserted

SQL> insert into fnd select rownum+11372,rownum+11372,rownum+11372 from dba_objects;
 
50070 rows inserted
 
Executed in 1.063 seconds
 
SQL> insert /*+append*/ into fnd select rownum+23772,rownum+23772,rownum+23772 from dba_objects;
 
50070 rows inserted
 
Executed in 0.781 seconds

SQL> insert into fnd select rownum+11372,rownum+11372,rownum+11372 from dba_objects;
 
50070 rows inserted
 
Executed in 1.063 seconds
 
SQL> insert /*+append*/ into fnd select rownum+23772,rownum+23772,rownum+23772 from dba_objects;
 
50070 rows inserted
 
Executed in 0.781 seconds
 
SQL> commit;
 
Commit complete
 
Executed in 0 seconds

SQL> select count(*) from fnd;
 
  COUNT(*)
----------
    250350
 
Executed in 0.094 seconds

SQL> select sum(num*num) from fnd;
 
SUM(NUM*NUM)
------------
477521168594
 
Executed in 0.125 seconds
 
SQL> select sum(num_float*num_float) from fnd;
 
SUM(NUM_FLOAT*NUM_FLOAT)
------------------------
         477522014000000
 
Executed in 0.078 seconds
 
SQL> select sum(ln(num*num)) from fnd;
 
SUM(LN(NUM*NUM))
----------------
5232311.70321546
 
Executed in 12.812 seconds
 
SQL> select sum(ln(num_float*num_float)) from fnd;
 
SUM(LN(NUM_FLOAT*NUM_FLOAT))
----------------------------
            5232311.70328157
 
Executed in 0.172 seconds
 
SQL> select sum(ln(num_double*num_double)) from fnd;
 
SUM(LN(NUM_DOUBLE*NUM_DOUBLE))
------------------------------
              5232311.70328157
 
Executed in 0.187 seconds

可见一般的运算number,binary_float,binary_double效率差不了多少,如果是复杂点的运算,特别是科学计算类的,binary_float,binary_double的效率就很明显.