Bandwidth of the Board System

During the DIY process, we often hear some old birds talking about the word "Bandwidth" and vaguely feel that this bandwidth is important, for example, the price of a high-bandwidth display is much higher than that of a General display. In fact, in a computer system, not only does the display and memory have the concept of bandwidth, but on a board, the concept of bandwidth is even more. It can be said that bandwidth is everywhere. Next, let's talk about some concepts about system bandwidth on the board.

Before understanding the concept of bandwidth, Let's first look at a formula: bandwidth = clock frequency X bus bits/8. We can see from the formula that, bandwidth is closely related to clock frequency and bus digits. The clock frequency is well understood. After years of experience from intel, we are familiar with the concept of frequency, and the bus bit is the width of the Data Bus) ,. This is in the video memory, listening more, such as those 64-bit graphics card, because the bandwidth is twice less than the-bit card, the efficiency is low, this is why they are often referred to as campus cards.

What is bandwidth? What is the significance of bandwidth? Simply put, bandwidth is the transmission rate, which refers to the maximum number of bytes transmitted per second (MB/S), that is, the number of megabytes processed per second. high bandwidth means the high processing capability of the system. In order to better understand the relationship between bandwidth, bit width, and clock frequency, let's give a more vivid example of how a worker processes parts. If one person is working at the same processing speed, the bandwidth is certainly not as high as the total number of processed parts. The bit width is like the number of workers. The clock frequency is equivalent to the speed of processing a single part. The wider the Bit Width, the higher the clock frequency, the larger the bus bandwidth, the obvious benefit.

The system bandwidth is shown on the board in the figure below. The main board usually has two relatively large chips. Generally, the one close to the CPU is called beiqiao, and the one far away from the CPU is called nanqiao. The function of beiqiao is to establish a communication interface between the CPU and memory and the video card. The bandwidth of the interfaces between them and beiqiao determines the memory and the video card performance to a large extent. Nanqiao is responsible for computer I/O devices, PCL devices and hard disks. Its bandwidth requirements are smaller than that of beiqiao. The connection bandwidth between the North and South bridges is generally called the bandwidth of the North and South bridges. With the development of computers to multimedia, the functions of nanqiao are becoming more and more powerful, and new requirements are also raised for the connection bus bandwidth between the North and South bridges. on Intel's 9x5 series motherboard, the bandwidth of the north-south bridge will grow from 266 Mb/s, which has been criticized for a long time to 2 Gb/s, resolving the bandwidth bottleneck between the North-South bridges.

In order to give readers a rational understanding of the bandwidth of the chipset, I made this table to indicate the bandwidth parameters of major chipset for your reference.

North Bridge part
return to the above part. The connection between the CPU and the North Bridge Chip is often referred to as the FSB (front side bus). It is a data channel between the CPU and the North Bridge. Its frequency directly affects the speed of CPU access to memory. For example, for P4 2.8c of 800 MHz bus frequency, according to the bandwidth calculation formula above, its bandwidth is 6.4 MHz x 64bit/8 = 400 Gb/s, while AMD's K7 processor's front-end bus is, similarly, the bandwidth is 3.2 Gb/s, Which is why dual channels are not very helpful to K7 processors, because a single channel ddr400 can meet the requirements of CPU access to memory, similarly, for the 800fsb P4 CPU, only the dual-channel DDR 400 can satisfy its appetite for bandwidth.
let's talk about the video card. All the friends who play the game know that when playing some big production games, the screen sometimes gets stuck. In fact, this is the problem of insufficient video card bandwidth. Specifically, this is a problem of insufficient memory bandwidth. As we all know, the current DMP interface on the road is the AGP 8x, and the frequency of the AGP bus is twice that of the PCL bus, that is, 66 MHz, it is easy to calculate that the bandwidth is 2.1 Gb/s. In the current environment, this bandwidth seems insignificant, even the most common ATI r9000 video memory bandwidth must reach 400 MHz X 6.4 bit/8 = Gb/s, not to mention other high-end graphics cards. Because of this, Intel uses the PCL-express bus in the latest 9x5 chipset to replace the old-fashioned AGP bus, compared with the traditional PCI and earlier computer bus shared parallel architecture, PCI Express uses point-to-point serial connections between devices. This allows each device to have its own dedicated connection and does not need to request bandwidth from the entire bus, at the same time, the data transmission speed can be easily raised to a high frequency by utilizing the serial connection features. In terms of transmission speed, since PCI Express supports two-way transmission mode, each device connected can use the maximum bandwidth. The bandwidth bottleneck encountered by AGP is also solved.
compare the bandwidth of each major bus, where Isa, PCL, AGP 1x, and AGP 2x have completely withdrawn from the historical stage because they are too old.

south bridge
As mentioned earlier, compared with the North Bridge with higher bandwidth requirements, the south bridge responsible for computer I/O equipment and Hard Disks, the bandwidth requirement is not very high, which is largely due to the performance constraints of the controlled devices. For example, the 7200-to-hard drive commonly used on PCs, even if the external transmission rate reaches 150 Mb/s, the internal transmission rate of the SATA hard drive is only 70 MB/S. For example, the frequency of the PCL Device controlled by nanqiao is only 33 MHz, naturally, the bandwidth is not high, so at present, nanqiao has not encountered any so-called bandwidth bottleneck.
to obtain more bus bandwidth during actual use of the computer, two methods are generally adopted based on the bandwidth calculation formula. One is to increase the bus speed, for example, Intel's P4 CPU and saiyang CPU are the best examples. One is the 400 bus, and the other is the 533/800 bus. The actual application performance is significantly different (of course, secondary cache is also an important factor ). Another common method is to increase the bus width. If the clock speed is the same, the bus width doubles, however, even though the clock descent along is the same as before, the data volume transmitted for each descent along is twice that of the previous one. This is at the same core, however, the video memory width is not the same as that on the video card.
by understanding the bandwidth of the system, I believe many friends have a new understanding of their computers and have a general understanding of the existence of bandwidth bottlenecks. But here, what I want to say is that a computer without any bottlenecks does not seem to exist at the moment, because a hard disk is enough to hold you down, without changing the existing computer structure, the bandwidth bottleneck of the system will not disappear. What we need to do is to deepen our understanding of hardware, it can correctly and rationally treat computer bandwidth bottlenecks and make full and reasonable use of existing computer resources.