The operation of the computer is carried out by the central processing unit. The most basic operations are arithmetic operations, namely+,-,*,/. The numerical values involved in arithmetic are different from those in pure mathematics, because "one foot pestle can't be used up in half a day". Only accurate numbers can be expressed inside the computer, and it is impossible to express infinite decimals. This is the first point: the numerical value of computer is different from that of human contact. The second point: the realization of computer operation is different from that of people. Inside the computer, two states are controlled by the on-off of the circuit, thus representing binary numbers; People use decimal numbers to express it. Therefore, it is decided that the computer can directly use the adder to complete the arithmetic operation of integers (in fact, non-negative integers), and further add a sign bit (whether the original code or the complement) to represent the operation of negative integers. So even the earliest CPU8088/8086 supports this type of operation. But the adder can't handle the operation of floating-point data, which is simply a numerical value with decimal point. For this kind of data, we must first have a good floating-point representation. In the end, the mantissa and sequential sign bits each occupy one bit, and then the mantissa of the remaining digits and the effective digits of the sequence are reasonably allocated. In other words, the binary scientific counting method is adopted. In operation, it is far more complicated than integer numerical operation, because the operation involves not only mantissa, but also order, and also the sign bits of mantissa and order, so the initial CPU has no floating-point operation ability (8088/8086, 80286, 80386SX). When floating-point operation is needed, it will be realized by CPU through software simulation, so it will be done when floating-point operation is carried out. In order to improve the computer performance, Intel Company designed a floating-point operation coprocessor 80387 for 80386SX CPU, which improved the operation speed when used together (80487SX with 80487). During the same period, Intel also produced CPU with integrated coprocessor: 80386DX and 80486DX. In the Pentium era, manufacturers integrated all coprocessors into CPU, which became an indispensable part, so there was no CPU with/without coprocessor. But the integer numerical operation is completed by adder, and the decimal numerical operation is completed by mathematical coprocessor integrated in CPU. Now all numerical operations are completed by the coprocessor, even if there is no separation.