Integrated Video Chipset Roundup - January 2001

Integrated Video: How it Works

The north bridge is perhaps the most convenient place to locate many system components, but most of all the video processor. This is due to one of the various functions which the north bridge controls which happens to be the AGP interface.

Since it is the north bridge that controls the interface for the vast majority of video cards out there, it only makes sense that any video integration occurs on this chip. By placing a video processor on the same silicon that controls the signals it gets and sends to the CPU, integrating a video processor on this chip not only makes sense but also reduces the traces and board space necessary to carry information to the offboard AGP graphics controller.

Although it is logical to place the graphics processor on the north bridge, we all know that there is more to a video card than just a processor. Not only must a video card process what it is sent, it must also store this information as well as provide a way to get it to the monitor. There are two different ways to attack the problem of temporary data storage.

The first approach involves placing separate RAM chips on the motherboard itself; in essence, physically soldering the same RAM chips found on current video cards onto the motherboard instead. From there, traces are made from the north bridge with integrated video to this onboard memory, providing a path for the data to travel. The speed of this path is dependent on the width of the path (how many bits it can send in one clock speed, 64-bit in the solutions we are looking at today) as well as the speed the data can travel. The total speed at which information can move from the integrated graphics chip to the external memory is called memory bandwidth and is measured by multiplying the bus width of the memory bus, times the clock speed of the memory, divided by 8 (the number of bits in a bite).

The use of separate SDRAM chips to provide a frame buffer for video storage is called local frame buffer mode and is typically avoided by manufacturers. This is because the addition of 4 to 32 MB of memory on the motherboard increases the base price of the motherboard. Since the goal of most of the integrated video solutions out there is to decrease cost, adding components that dramatically increase the cost of a motherboard are avoided for the most part.

The road more commonly traveled in the integrated video market involves a frame buffer solution called unified memory mode. In a unified memory mode configuration, a systems own memory (RAM) is used as a frame buffer for the video card. A certain amount of system SDRAM is set aside to act solely as the integrated video's frame buffer. Just as in a local frame buffer solution, the speed at which data travels to and from the system memory is dependent on both the bus width and the memory speed. In the case of unified memory solutions, the bus width is always 64-bits wide (for all integrated solutions out there currently, anyway) and the speed is from 100 MHz to 133 MHz, or the speed that the system memory performs at. Although this may sound like the memory bandwidth of this setup would be similar to the memory bandwidth of a local frame buffer solution (since both have a 64-bit bus and typically 133 MHz memory), the memory bandwidth of a shared memory buffer solution actually proves to be much less.

The reason for this is simple. Unlike a local frame buffer solution that has its own path to the frame buffer, unified memory solutions share a path with a computer's bus, a path already crowded as it is. Sharing a path with the same system that is already loaded down by carrying information from the CPU to the system memory and back, the actual memory bandwidth in a unified memory solution is far less than the ideal 1.06GB/s of theoretical bandwidth available in a system with a 133 MHz front side bus.

It is much cheaper to rely on a user's system memory to serve as a frame buffer, as no RAM chips need to be physically added to the motherboard; the system is total dependent on memory added after the sale of the motherboard. For this reason, the vast majority of integrated video systems out there work in shared memory mode, saving production costs but hampering performance.

As for the other parts of a video card, all elements are contained on the motherboard itself. Filters used to provide a clean video signal, as well as the video BIOS and other necessary components, are placed where ever convenient on the motherboard design. The 15-pin VGA port is almost exclusively positioned in the location of one of the two serial ports found on the ATX form factor. This is a convenient location for the port, as both the 15-pin VGA port and the 9-pin serial port are essentially the same size.

Apart from the aforementioned components, many of the items found on a standard video card are actually power regulating components that are not necessary in an integrated solution.

Now that we know a bit about the types of integrated video systems out there as well as how they work, lets review what solutions are currently being offered before we see which one performs best.