SQL Server datatype precision - Neo, what is real?

Question

SQL Sever 2000 documentation:

Is a floating point number data with the following valid values: –3.40E + 38 through -1.18E - 38, 0 and 1.18E - 38 thr

Answer 1

Your answer is on that same page you linked:

For FLOAT data type, the n is the number of bits used to store the mantissa in scientific notation

24 bits of mantissa gives you (approximately) 7 decimal digits of precision (because 2^24 ~= 10^7).

edit to add:

Notice that everyone keeps saying 'approximately' - this is for a reason :)

Binary floating point numbers and decimal literals do not necessarily play together in an intuitive manner. For background read What Every Computer Scientist Should Know About Floating-Point Arithmetic. Also note that saying 'approximately 7 decimal digits of precision' is not incompatible with being able to store a value with more than 7 significant figures! It does mean however that this datatype will be unable to distinguish between 0.180000082 and 0.180000083, for example, because it isn't actually storing the exact value of either:

declare @f1 real
declare @f2 real

set @f1 = 0.180000082
set @f2 = 0.180000083

select @f1, @f2

select @f1 - @f2

---

------------- -------------
0.1800001     0.1800001

(1 row(s) affected)


-------------
0

(1 row(s) affected)

---

The fact is that real is the same as float(24), a binary floating point number with 24 bits of mantissa, and I don't believe there's a way to change this. Floating-point types are in general not a good choice if you want to store exact decimal quantities.