What is AGP?

Designed for Pentium II based motherboards, AGP will, among many other benefits, deliver a peak bandwidth that is 4 times higher than the PCI bus using pipelining, sideband addressing, and more data transfers per clock. It will also enables graphics cards to execute texture maps directly from system memory instead of forcing it to pre-load the texture data to the graphics card's local memory. AGP is based on the PCI 2.1 standard which calls for a 66MHz PCI bus speed.

The result of AGP is a much smoother frame rate and the ability to display 3D graphics and video that is many times more realistic and much a higher quality than ever before found a PC. Though they are obviously biased, Intel makes the claim that, "Now PC users can experience the type of dazzling 3D graphics and video previously found only on workstations costing $20,000 or more!"

One thing is for sure about that comment - the price for this new AGP technology is a mere fraction of a $20,000 workstation.

Though AGP is geared towards the Pentium II and it's Dual Independent Bus (DIB) architecture, you don't need a Pentium II processor to take advantage of the new technology, since the use of AGP doesn't depend on the CPU type. In fact, VIA has a new chipset due out soon for Socket 7 motherboards that supports AGP.

Data Transfers

The current PCI bus supports a data transfer rate up to 132 MB/s, while AGP (at 66MHz) supports up to 533 MB/s! AGP attains this high transfer rate due to it's ability to transfer data on both the rising and falling edges of the 66MHz clock, and through new design advances that have made data transfer modes more efficient.

AGP Architecture Diagram

Direct Memory Execute (also knows as DIME) is probably the most important feature of AGP. AGP graphic chips have the capability to access main memory directly for the complex operation of texture mapping. AGP provides the graphics card with two methods of directly accessing texture maps in system memory: pipelining and sideband addressing. In pipelining, AGP makes multiple requests for data during a bus or memory access. PCI makes one request, and does not make another until the data it requested has been transferred.

AGP "DIME" Diagram

Company Support

It's great that Intel designed the AGP bus, but who's going to implement it? The following is a list of companies who are members of the AGP Implementors Forum. You can be sure that you will see these companies use AGP in their upcoming products:

In summary, the benefits of AGP include:

1. Peak Bandwidth 4x the PCI bus, and higher sustained rates via sideband and pipelining.
2. Direct Memory Execute of textures.
3. Reduced Contention with the CPU and I/O devices for bus and memory access. The PCI bus serves disk controllers, LAN chips, and possibly video capture. AGP operates concurrently with, and independent from, most transactions on PCI. Further, CPU accesses to system RAM can proceed concurrently with the graphics chip's AGP RAM reads, because of so-called out-of-order and queuing hardware support in the chip set. So in spite of heavy access from the graphics chip, there should be no audio breakup or other CPU degradation.
4. An "extra port" to the graphics chip for memory access, so it can concurrently read textures from AGP memory while reading/writing Z-values and pixels from local memory. Using the bandwidths from Figure 2, the graphics chip gets 1.3 GB/s peak by using both ports simultaneously, versus "only" 0.8 GB/s from the local RAM.
5. Allowing the CPU to write directly to shared system AGP memory when it needs to provide graphics data, such as commands or animated textures. Generally the CPU can more quickly access main memory than it can graphics local memory via AGP, and certainly faster than via the PCI bus.

Be looking for a review of a motherboard with Intel's new 440LX "AGPset", as well as a typical AGP graphics card. It's certainly too early to say just how much of a speed increase that AGP will bring in the real world. Intel benchmarked AGP vs. PCI using ZD Lab's 3D Winstone. Here is a chart from Intel's site which shows their results:

The only problem with these benchmarks is that the actual test that was used was a "Large Texture Scene", which includes texture data up to 6 MB. Thus, of course AGP is going to win out over PCI since the large amount of data can be read directly from the system memory rather than having to be loaded into the graphics card's memory.

So this is good news to some extent: we should expect much better performance at high resolution graphics that require large amounts of memory. It's unknown as to whether low resolution graphics that require small amounts of memory will benefit as much from AGP.