14 DAY TRIAL // And if you think about doing a SLI configuration between two PCIE 2.0 cards that will surely translate into a mighty 2x32x setup. Or 4x16x if you prefer Quad SLI configurations. Anyway, the PCIE 2.0 standard is on its way of becoming the next-generation connector that will be integrated in the motherboards. And in case PCIE 2.0 sounds too heavy at the moment, let's make some light.
First I will point out that PCIE has a fixed frequency of 100MHz and a single serial link (x1) provides 250MB/s of throughput. Multiply it with 16x and you'll get 4GBs/s of transfer for your 16x capable video card. In the case of 2x16x SLI setups, the bandwidth doubles to 8GBs/s if the motherboard can provide 2 16x independent channels.
However, things are not that simple. The PCIE transfer can also be measured in Giga Transactions per second (or GT/s). In this case, a fully capable 16x PCIE slot can deliver 2.5GT/s. While at first this transfer rate seems to have nothing in common with the GBs that flow through the connector, it actually does. The term "transaction" refers to the flow of 8-bit words through the PCIE interface. But actually 10 bits are transferred every time because of the need to introduce error checking algorithms. So for every 8 bits of data, at least 10 physical bits are transferred. The numbers can go even higher because of the overhead that appears on some situations.
PCIE 2.0 is said to up the transfer rate to 5GT/s producing a single 32x connector with as much as 8GB/s of physical throughput. Enlarge the reasoning to SLI setups and you'll get a 2x32x configuration capable of delivering about 16GB of data every second. And that's pretty impressive when you look at the numbers. However, real life seems to talk other languages. You all know that the difference between 8x and 16x VGA configurations is about 5-10% and the performance leap should be at most 10% when the motherboard producers will switch to 32x PCIE slots. The advantages of the 2.0 slots will only be felt if VGA cards that need a lot more bandwidth appear. Until then, I'm pretty happy with the classical ones.