PCI Express Versions 2 and 3: Not Just More Speed But More Watts

Having covered the bandwidth of PCI Express Version 1.x and 2.0, we look at the electrical power they can deliver to a graphics card, then bring version 3 into the mix.

The increasing performance of graphics cards has not only put increasing demands on the bandwidth they require. The performance boost has come at the expense of using more power, with some top-of-the-line cards currently drawing over 200 watts. That power has to come from somewhere.

PCI Express 1 and 2 deliver 75 watts through the slot itself. Since that won’t even power a mid-range card, extra power is drawn directly from the computer’s power supply unit. PCI-E developed alongside specifications to get the power needed via a connector from the power supply. This was originally a single, 6-pin connector that was able to deliver 75 watts, and was eventually joined by a second identical connector, for a total of 225 watts.

The current standard calls for an 8-pin connector, capable of delivering 150, which can be supplemented with one of the older, 75 watt connectors. Add the 75 watts from the slot and the total maximum is 300 watts. While this is more than the average user will ever need, there are graphics cards, and people who buy them, that will use this much power. ATI’s Radeon 4870 X2, for instance, will draw almost 270 watts at full tilt. Some entire computers, albeit very slow ones, run on 250 watts or less altogether. You can learn more about choosing a power supply here.

Instead of resting on their laurels, the PCI-SIG (Peripheral Component Interconnect Special Interest Group), which controls PCI specifications, is working on PCI Express 3. This is likely because many of PCI-SIG’s members sell graphics cards, and they want to make sure you won’t hesitate to buy one because you don’t have the bandwidth.

PCI-E 3.0 will double PCI-E 2.0 speeds to 1 GB/s per lane, or 16 GB/s for a x16 graphics card slot. Part of the increase will come from increasing the signaling rate, or just turning up the clock because manufacturing improvements allow for it. The rest of the increase comes from leaving behind 8b/10b encoding. Like a SATA connection, PCI-E 1 and 2 would add 2-bits to every byte: PCI-E 3 won’t.

On the wattage side, little is known. There are, however, rumors that PCI-E 3 will include support not only for 300 watt cards, but will physically be able to support a 1.5 kilo card. That’s support in terms of holding it up without dropping it, and kilo in terms of grams. Such a card would be behemoth, being an unprecedented three slots wide. Low to lower-midrange cards use take up one slot. Upper-midrange cards take up two with bulky cooling. A triple-wide boggles the mind at first, but starts to make sense if one considers a double GPU card with a lot of cooling.

PCI-E 3 will be backwards compatible. You will be able to put a version 1, 2, or 3 card in any PCI-E slot (though getting enough watts to a top end-card, and keeping it from falling out, may be a problem) and vice-versa. The combination will be limited to the slower of the two though.

Calm down. PCI-E 3 won’t be in stores until 2010. Even then, the best available graphics hardware will take some time to make full use of the slot, and the midrange will take years after that to need all of PCI-E 3’s bandwidth. Plus, the backwards compatibility means a graphics card can survive a motherboard upgrade and vice versa.

PCI-E 1 might be getting a little long in the tooth, but if you are planning an upgrade, don’t hold off on PCI-E 3’s account. PCI-E 2 is the way to go. For more tips on choosing a motherboard, see this article.