The Quick PCI-Express 2.0 Guide

About a year ago, the PCI Special Interest Group (PCI-SIG) released version 2.0 of the PCI-Express (PCIe) standard. We are now beginning to see hardware on the market that implements the new standard. What does this mean for you?

Recap of PCI-Express basics

In order to understand what has changed, it helps to know a little about how PCIe works.

PCIe is a point-to-point interconnect intended to replace both PCI (a general-purpose system bus) and AGP (a bus used for graphics cards). It is based on the concept of lanes, which work by analogy to lanes on a highway. A single physical slot that you plug a card into may have between 1 and 32 lanes, in practice always a power of 2 (1, 2, 4, 8, 16). Although x32 slots can theoretically exist, we have no good evidence that anyone has ever built one into a commercial product.

Just like a highway, more lanes means the ability to carry more traffic. Each PCIe lane can carry up to a fixed maximum amount of data; for PCIe 1.0a and 1.1, that amount is 250 MB/s in each direction.

The nomenclature for a PCIe slot is to refer to it based on the number of lanes it has. A slot with 1 lane is a x1 (spoken "by one") slot. This is meant to be similar-but-different to AGP's 1x, 8x, etc. system.

The great source of confusion

PCIe slots can be described by two properties: their version number and the number of lanes that they support. Because the concept of PCIe lanes is not generally well-understood by the public, this has created confusion. So to be clear: PCIe 2.0 is not replacing PCIe x16. That is not how it works.

New motherboards that support PCIe 2.0 will continue to have x16 slots. Those slots now have beefier capabilities, but they are still x16 slots and they are still intended primarily for video cards.

It is a computing tradition that all concepts, whenever possible, must be explained through a complicated car analogy. So here is one: Suppose you want to buy a car (motherboard) with a V6 engine (PCIe x16 slot). You might buy something from the older 2007 model year (PCIe 1.x), or you might buy something from the newer 2008 model year (PCIe 2.0). Either way, you are still buying a V6. The 2008 model happens to perform better, but it is still a V6.

What has changed

The major change in the new revision of the standard is increased bandwidth. The older PCIe 1.x standards let each lane transmit up to 250 MB/s of data; PCIe 2.0 ratchets this up to 500 MB/s. This means that a x16 slot, such as the one most graphics cards plug in to, can deliver up to 16 GB/s of bandwidth (that is both directions combined). Twice as much, in other words, as older versions of the spec.

The power spec has also been increased. A new version of the power supply connector that is used for graphics cards that need more power than the slot can deliver (sometimes called a PEG connector, for PCI-Express Graphics) has been introduced, changing from 6 pins to 8. A single x16 card may now draw up to 300 W of power (75 W from the slot itself, 150 W from an 8-pin PEG connector, 75 W from a second PEG connector), up from 225 W (75 W from the slot, 75 W each from 2 6-pin PEG connectors) or originally 150 W (75 W from the slot, 75 W from a 6-pin PEG connector).

The links are also now smarter. They can negotiate a new link width (number of lanes) or speed (whether to use the clock speed of PCIe 1.x or 2.0) on the fly. Power requirements can be renegotiated on the fly. The operating system can be informed of such changes.

Backward compatibility: new motherboard, old card

PCIe 2.0 is 100% backward compatible. Any PCIe 1.x card should work in any PCIe 2.0 motherboard. The link will have to clock down to the slower PCIe 1.x speed. But since that is as fast as the card is designed to communicate anyway, no performance will be lost.

If you have a motherboard that is somehow not compatible with older cards, contact the company that made your motherboard because they screwed up.

The new 8-pin PEG connectors are not necessarily physically compatible with older 6-pin receptacles. Many power supplies use "breakaway" connectors (where 2 pins can be removed fronm the connector), or provide both types of connectors, to work around the issue.

Forward compatibility: old motherboard, new card

PCIe 2.0 is intended to be 100% forward compatible. Any PCIe 2.0 card should work in any PCIe 1.x motherboard. Again the link speed will be limited by the slower part, and will have to clock down to the slower PCIe 1.x speed. Because the card expects to have more bandwidth available, this could cause a loss of performance (as of this writing that loss is purely theoretical, as no commercial cards can saturate a x16 PCIe 1.x link).

There are scattered anecdotal reports of some very early PCIe motherboards, boards that implemented the PCIe 1.0a spec, not playing nice with new PCIe 2.0 cards. Again, this comes down to a violation of the specification on the part of either the motherboard or card. If you are stung by such a problem, contact the companies that made your motherboard and video card.

Because of the change in power connectors, new cards are not guaranteed to be compatible with older power supplies. In particular, a card might draw more power than the power supply (with all of the other hardware in the system) can comfortably supply. Additionally, adapters may be needed for very high-power cards; although the older 6-pin PEG connectors will fit into newer 8-pin receptacles, they cannot supply as much power. Such cards might require the use of an adapter that turns Molex connectors into an 8-pin PEG connector, for example.

Further reading

PCI-SIG overview of PCIe 2.0

PCI-SIG PCIe 2.0 Frequently Asked Questions

Ars Technica overview

Anandtech's overview

  • RSS feed StumbleUpon Digg Yahoo! My Web 2.0