Processor operation number of bits

The CPU's bit width has no less impact on CPU performance than the main frequency. Bit width refers to the data bandwidth of the microprocessor executing instructions at one time. The addressing width of the processor has grown rapidly. The industry has used 4, 8, and 16-bit addressing to reach the current mainstream 32-bit, and 64-bit addressing floating-point arithmetic has gradually become the mainstream product of the CPU.

Due to the limitations of virtual and real memory sizes, the current mainstream 32-bit CPUs have a serious flaw in the performance execution mode: when faced with a large number of data streams, 32-bit registers (Note: To process data, temporarily store results, or Indirect addressing and other actions, each processor has some built-in memory, these memory can be accessed without delay is called "registers", each register is the same size) and instruction set Can't make corresponding processing calculations in time.

A 32-bit CPU can only handle 32 bits at a time, that is, 4 bytes of data; and a 64-bit CPU can handle 64 bits or 8 bytes of data at a time. If we edit the 128-bit instruction in 16-bit, 32-bit, and 64-bit units, respectively, the old 16-bit CPU (such as the Intel 80286 CPU) requires 8 instructions, and the 32-bit CPU requires 4 instructions. A 64-bit CPU requires only two instructions. Obviously, with the same operating frequency, 64-bit CPUs are faster than 16-bit and 32-bit processors.

You can compare the 32-bit and 64-bit CPUs in the figure. The number of 64-bit code streams does not change. The width varies with the width of the instruction code; the data stream width doubles. Although the theoretical amount of data handled by a 64-bit system is twice that of a 32-bit system in one clock cycle, there is usually a gap between theory and reality.

It should be noted that the CPU not only requires a register with a wide enough bit width but also a sufficient number of registers to ensure large amounts of data processing. Therefore, in order to accommodate more data, registers and internal data channels must also be doubled, so the number of registers in a 64-bit CPU is typically twice that of a 32-bit CPU.

However, although the number of register bits has increased, the instruction registers that are executing the instructions are all the same, that is, the data stream is doubled and the instruction stream is unchanged. In addition, increasing the number of data bits can also increase the dynamic range. In the commonly used decimal system, only up to 10 integers (in the case of a single digit) can be obtained. This is because there are only 10 different symbols in 0 to 9 to indicate the corresponding meanings, and it is necessary to represent 10 or more numbers. Add one digit, two digits (00-99) can represent 100 numbers.

The calculation formula for the decimal dynamic range can be derived: DR=10n (n indicates the number of digits). In the binary system, correspondingly we can get the formula: DR=2n, then the 32 bits currently used can reach 232=4.3×109. After upgrading to 64 bits, it can reach 264=1.8×1019. The dynamic range is expanded. 4.3 billion times.

Tip: Enlarging the dynamic range can improve the accuracy of the data in the register to some extent. For example, when using a 32-bit system to process meteorological simulation tasks, when the processed data exceeds the maximum dynamic range that 32 bits can provide, the system will appear such as Overflow (more than the maximum positive integer) or Underflow (below the minimum Negative integer error, so that the data in the register cannot be guaranteed accurate.

In addition to the computing power, the advantage of the 64-bit CPU compared to the 32-bit CPU is also reflected in the system's memory control. Since the address uses a special integer, an ALU (Arithmetic Logic Operator) and register of the 64-bit CPU can handle larger integers, ie, larger addresses.

The traditional 32-bit CPU has a maximum address space of 4 GB, which makes many large-scale data processing programs that require large-capacity memory to be stretched at this time, thus creating a bottleneck for operating efficiency. The 64-bit processor can theoretically reach 18 million TB (1TB = 1024GB), which will completely solve the bottleneck encountered by 32-bit computing systems.

Of course, the 64-bit addressing space also has certain disadvantages: the memory address value becomes twice as many as the number of bits, so that the memory address will occupy more space in the cache, and other useful data cannot be loaded into the cache. , which caused a certain degree of decline in overall performance.