Graphics Caught Up to PCI-E 1.1 Bandwidth
Before we get going, we’ll point out we are referring to theoretical maximum bandwidth. This means the speed that could be achieved in a perfect world, if no component had to spend time thinking about what to do or waiting for something else to get out of the way. A remote control that works ‘up to 100 feet’ might assume you are in a lead chamber under vacuum to get that number. Theoretical bandwidth is that kind of never going to happen kind of number, but it is important in terms of illustrating a technology’s potential.
A single lane, x1 PCI Express slot offers 250 MB/s of bandwidth, almost twice the 133 MB/s shared by all PCI slots. Plus PCI-E is full duplex, allowing that much data to be sent in each direction at once. Arguing that that doubles the bandwidth, requires assuming the card and computer are always sending each the same amount of data, so that isn’t generally done. But it is obviously faster than one way at a time, called half-duplex.
By using a different number of lanes, different PCI-E slots offer different speeds. One just multiplies 250 by the number of lanes. That means an x16 graphics slot will offer 4 GB/s. That sounds like a lot, and in 2004, it was.
Many chose to stick with AGP: we covered the current lack of popularity of that standard in the previous article. They are either making do with a limited and dwindling selection of low-end graphics cards, or have already forked over the cash on a new mobo.
Graphics cards can already take advantage of PCI-E 2.0 bandwidth when dealing with detailed textures, or on dual GPU cards, where two graphics processors share a card and slot. If you are running multiple high-end cards, you will definitely need to pay attention to how many lanes they all have.