Comment Added by : William Fraser

Comment Added at : 2015-06-07 16:37:52

Comment on Tutorial : float vs double data types in Java By Baski
Actually, the given accuracy just barely exceeds what is possible in a float. A float is stored as: 1 sign bit, 23 mantissa bits, 1 sign bit for the exponent, and 7 bits for the exponent value.

Since the largest value that can be stored in 23 bits is 8388608, two consecutive floats in this region are:

8380398/131072 = 63.93736267
8380399/131072 = 63.9373703

Meaning that it can just barely distinguish between 63.93736 and 63.93737, which is still MUCH better than the human eye!

View Tutorial