84 Computer Architecture and Organization
Figure 4.26 Number density in scientific notation
already indicated that the decimal part within 4-bit representation would cover a range between 0 and 9.
It cannot cover a range upto 15 as it does not complete the maximum 2-digit value, i.e., 99. However,
it offers accuracy upto rst place of decimal within the range covered by it. Now, let us consider the
second representation (Figure 4.25 ). The range in this case would be from −15.5 to +15.5. As it may be
readily observed, the available range is extended by adopting the scienti c notation using exponential
system. Therefore, it is preferable to adopt this scheme for representing all types of decimal numbers
within computer memory.
4.7.7 Accuracy
As we can observe from Figure 4.26 , lesser is the value of the exponent, lesser would be the spacing
of the representative numbers. Therefore, between 1 and 2, we observe a spacing of 0.0625, which is
increased to 0.125 between the range from 2 to 4. Between 4 and 8, this spacing (or interval) increases
to 0.25 and, nally between 8 and 15.5, it becomes 0.5. Graphically, this is presented at the bottom of
Figure 4.26 . This indicates one biggest limitation of our notation scheme that we are unable to represent
any number. For example, we are unable to represent 14.25 through our 8-bit scheme. We have to round
it off to either 14.0 or 14.5, so that it may be expressed through our scheme. Therefore, some approxi-
mation as well as rounding off are necessary for representing oating-point numbers using scienti c
notation scheme.
Another point must be evident to the reader that the range of numbers would increase if we expand
the size (available bits) of the exponent. However, if we have to increase the accuracy by decreasing the
space between two successive numbers, then the bits reserved for mantissa must be more. Next comes
the case of trade off. In computers we generally get the word length of 32 bits or 64 bits. Generally,
for higher accuracy, we go for double precision of 64-bit representation. For less accurate cases, 32-bit
representation is taken as suf cient.
M04_GHOS1557_01_SE_C04.indd 84M04_GHOS1557_01_SE_C04.indd 84 4/29/11 5:04 PM4/29/11 5:04 PM