|
|
It's PCI for GraphicsMar 06 '01 Write an essay on this topic.
Popular Products in PC Desktops
The Bottom Line A graphics expansion slot that runs faster than PCI with fewer conflicts.
To understand AGP (Advanced Graphics Port), you must first understand PCI (Peripheral Component Interface). The PCI bus is how data is transferred from peripheral devices (modems, network cards, sound cards, video cards, etc) to the processor (and directly to memory with DMA, but that's another story). In essence, the PCI bus is a series of 124 wires, or "pins," that carry electricity. 32 of these pins carry data while the rest supply power and ground or special signal information to the PCI device. Strangely enough, with 32 pins for data, a PCI device can send 32 bits of data at one time. There is a little clock chip that tells it how often it can send data so that it stays nicely timed at 33 million transfers per second, or 33 MHz. There are versions of the PCI bus that transfer 64 bits at a time, or that operate at 66 MHz, but they work the same way. There is one complication, though. The same bus services several PCI devices. What if two devices want to transfer data at the same time? They have to resolve this conflict or things break. This is where we get Interrupt Requests (IRQs). IRQs are how devices decide to take priority over each other for the PCI bus. Device A will be transferring some data when device B decides that it urgently needs to send something. Device B sends some electricity along one of those special signal pins to ask for control of the bus. The time it takes to get control depends on how its IRQ ranks and what other interrupts are currently being requested. This delay is extremely short(a few millionths of a second at the most) and generally doesn't affect performance at all. One task that was affected by this performance hit was graphics. Not only did the graphics card really hate being told to wait, it was using the PCI bus so much that all the other devices were getting fed up with the bandwidth hog. This is where AGP comes into play. The AGP specification was initially almost identical to the PCI specification from a logical standpoint. There were differences like electrical requirements and how the pins were arranged but it performed the same function. The major difference was that it was the only device on the bus, so there was no need to process interrupts or wait for other devices. While this may not have affected most people, it was nice for people who had lots of PCI devices, or other PCI devices that used a lot of bandwidth (SCSI controllers, heavily used NICs, etc). Then came 2x AGP. The clock chip stayed at 33 MHz, but now instead of just sending one set of 32 bits per clock cycle, it was sending two. 4x AGP made this even better by sending four sets of data per clock cycle. Now 8x AGP is in the works which should make for some truly amazing bandwidth. There are a number of other enhancements that have been made to the AGP bus, such as direct access to memory, voltage modifications and logical proximity to the processor, some of which are fairly important themselves, but these will wait for another article.  |
| Read all comments (1)|Write your own comment |
