How PCI Express Works

by Tracy V. Wilson

Browse the article How PCI Express Works

Introduction to How PCI Express Works

Peripheral Component Interconnect (PCI) slots are such an integral part of a

computer's architecture that most people take them for granted. For years, PCI has
been a versatile, functional way to connect sound, video and network cards to a
motherboard.

But PCI has some shortcomings. As processors, video cards, sound cards and
networks have gotten faster and more powerful, PCI has stayed the same. It has a
fixed width of 32 bits and can handle only 5 devices at a time. The newer, 64-bit PCI-
X bus provides more bandwidth, but its greater width compounds some of PCI's other
issues.

A new protocol called PCI Express (PCIe) eliminates a lot of these shortcomings,
provides more bandwidth and is compatible with existing operating systems. In this
article, we'll examine what makes PCIe different from PCI. We'll also look at how
PCI Express makes a computer faster, can potentially add graphics performance, and
can replace the AGP slot.

Photo courtesy Consumer Guide Products

Thank You
Thanks to Joshua Senecal for his assistance with this article.

High-Speed Serial Connection

In the early days of computing, a vast amount of data moved over serial connections.
Computers separated data into packets and then moved the packets from one place to

1
another one at a time. Serial connections were reliable but slow, so manufacturers
began using parallel connections to send multiple pieces of data simultaneously.

It turns out that parallel connections have their own problems as speeds get higher and
higher -- for example, wires can interfere with each other electromagnetically -- so
now the pendulum is swinging back toward highly-optimized serial connections.
Improvements to hardware and to the process of dividing, labeling and reassembling
packets have led to much faster serial connections, such as USB 2.0 and FireWire.

PCI Express is a serial connection that operates more Sizing Up

like a network than a bus. Instead of one bus that Smaller PCIe cards will fit into
handles data from multiple sources, PCIe has a switch larger PCIe slots. The computer
that controls several point-to-point serial connections. simply ignores the extra
connections. For example, a x4
(See How LAN Switches Work for details.) These card can plug into a x16 slot. A
connections fan out from the switch, leading directly x16 card, however, would be too
to the devices where the data needs to go. Every big for a x4 slot.
device has its own dedicated connection, so devices
no longer share bandwidth like they do on a normal bus. We'll look at how this
happens in the next section.

PCI Express Lanes

When the computer starts up, PCIe determines which devices are plugged into the
motherboard. It then identifies the links between the devices, creating a map of where
traffic will go and negotiating the width of each link. This identification of devices
and connections is the same protocol PCI uses, so PCIe does not require any changes
to software or operating systems.

2
Each lane of a PCI Express connection contains two pairs of wires -- one to send and
one to receive. Packets of data move across the lane at a rate of one bit per cycle. A
x1 connection, the smallest PCIe connection, has one lane made up of four wires. It
carries one bit per cycle in each direction. A x2 link contains eight wires and transmits
two bits at once, a x4 link transmits four bits, and so on. Other configurations are x12,
x16 and x32.

3
 Scalable PCI Express slots.

PCI Express is available for desktop and laptop PCs. Its use may lead to lower cost of
motherboard production, since its connections contain fewer pins than PCI
connections do. It also has the potential to support many devices, including Ethernet
cards, USB 2 and video cards.

Two by Two
The "x" in an "x16" connection stands for "by." PCIe connections
are scalable by one, by two, by four, and so on.

But how can one serial connection be faster than the 32 wires of PCI or the 64 wires
of PCIx? In the next section, we'll look at how PCIe is able to provide a vast amount
of bandwidth in a serial format.

PCI Express Connection Speeds

The 32-bit PCI bus has a maximum speed of 33 MHz, which allows a maximum of
133 MB of data to pass through the bus per second. The 64-bit PCI-X bus has twice
the bus width of PCI. Different PCI-X specifications allow different rates of data
transfer, anywhere from 512 MB to 1 GB of data per second.

4
 Devices using PCI share a common bus, but each device using
PCI Express has its own dedicated connection to the switch.

A single PCI Express lane, however, can handle 200 MB of traffic in each direction
per second. A x16 PCIe connector can move an amazing 6.4 GB of data per second in
each direction. At these speeds, a x1 connection can easily handle a gigabit Ethernet
connection as well as audio and storage applications. A x16 connection can easily
handle powerful graphics adapters.

How is this possible? A few simple advances have contributed to this massive jump in
serial connection speed:

• Prioritization of data, which allows the system to move the most important
data first and helps prevent bottlenecks

5
 • Time-dependent (real-time) data transfers
• Improvements in the physical materials used to make the connections
• Better handshaking and error detection
• Better methods for breaking data into packets and putting the packets together
again. Also, since each device has its own dedicated, point-to-point connection
to the switch, signals from multiple sources no longer have to work their way
through the same bus.

Slowing the Bus

Interference and signal degradation are common in parallel
connections. Poor materials and crossover signal from nearby
wires translate into noise, which slows the connection down. The
additional bandwidth of the PCI-X bus means it can carry more
data that can generate even more noise. The PCI protocol also
does not prioritize data, so more important data can get caught in
the bottleneck. Using the Accelerated Graphics Port (AGP) slot
for video cards removes a substantial amount of traffic, but not
enough to compensate for faster processors and I/O devices.

PCI Express and Advanced Graphics

We've established that PCIe can eliminate the need for an AGP connection. A x16
PCIe slot can accommodate far more data per second than current AGP 8x
connections allow. In addition, a x16 PCIe slot can supply 75 watts of power to the
video card, as opposed to the 25watt/42 watt AGP 8x connection. But PCIe has even
more impressive potential in store for the future of graphics technology.

Photo courtesy Consumer Guide Products

PCI Express video card

6
 Photo courtesy Consumer Guide Products
AGP 8x video card

With the right hardware, a motherboard with two x16 PCIe connections can support
two graphics adapters at the same time. Several manufacturers are developing and
releasing systems to take advantage of this feature:

• NVIDIA Scalable Link Interface (SLI): With an SLI-certified motherboard,

two SLI graphics cards and an SLI connector, a user can put two video cards
into the same system. The cards work together by splitting the screen in half.
Each card controls half of the screen, and the connector makes sure that
everything stays synchronized.

Photo courtesy NVIDIA

NVIDIA SLI link card

7
 • ATI CrossFire: Two ATI Radeon® video cards, one with a "compositing
engine" chip, plug into a compatible motherboard. ATI's technology focuses
on image quality and does not require identical video cards, although high-
performance systems must have identical cards. Crossfire divides up the work
of rendering in one of three ways:
o splitting the screen in half and assigning one half to each card (called
"scissoring")
o dividing up the screen into tiles (like a checkerboard) and having one
card render the "white" tiles and the other render the "black" tiles
o having each card render alternate frames
• Alienware Video Array: Two off-the-shelf video cards combine with a Video
Merger Hub and proprietary software. This system will use specialized
cooling and power systems to handle all the extra heat and energy from the
video cards. Alienware's technology may eventually support as many as four
video cards.

Photo courtesy NVIDIA

Two video cards running parallel

Since PCI, PCI-X and PCI Express are all compatible, all three can coexist
indefinitely. So far, video cards have made the fastest transition to the PCIe format.
Network and sound adapters, as well as other peripherals, have been slower in
development. But since PCIe is compatible with current operating systems and can
provide faster speeds, it is likely that it will eventually replace PCI as a PC standard.
Gradually, PCI-based cards will become obsolete.

8
Bandwidth
PCI Express in all it's flavors: 1x, 2x, 4x, 8x, 16x and 32x all have much greater bandwidth
than basic PCI.

Common Buses and their Max Bandwidth

PCI 132 MB/s
AGP 8X 2,100 MB/s
PCI Express 1x 250 [500]* MB/s
PCI Express 2x 500 [1000]* MB/s
PCI Express 4x 1000 [2000]* MB/s
PCI Express 8x 2000 [4000]* MB/s
PCI Express 16x 4000 [8000]* MB/s
PCI Express 32x 8000 [16000]* MB/s
IDE (ATA100) 100 MB/s
IDE (ATA133) 133 MB/s
SATA 150 MB/s
Gigabit Ethernet 125 MB/s
IEEE1394B [Firewire] 100 MB/s

* Note - Since PCI Express is a serial based technology, data can be sent over the bus in two
directions at once. Normal PCI is Parallel, and as such all data goes in one direction around
the loop. Each 1x lane in PCI Express can transmit in both directions at once. In the table the
first number is the bandwidth in one direction and the second number is the combined
bandwidth in both directions. Also please note that in PCI Express bandwidth is not shared
the same way as in PCI, so there is less congestion on the bus.

Increased bandwidth can be equated into increased system performance. We've long known
that to get the most out of your processor you need to get as much information into it as
possible, as quickly as possible. Chipset designers have consistently addressed this by
increasing Front Side Bus speeds. The problem with this is that front side bus speed
increases the speed of transfer between the memory and CPU but often you've got data that's
coming from other sources that needs to get to the memory or CPU like drives, network
traffic, video, etc. PCI Express addresses this problem head on by making it much faster and
easier for data to get around the system.

9
Physical Differences: PCI Express [ PCI-E ] vs AGP vs PCI

Currently, the most common use for

PCI Express is Video. On the
graphic at right you can see the
physical differences between the
cards.

The connector on PCI Express

video cards will always start out with
a small piece not directly joined with
the rest of the slot, and then a
notched piece not directly joined
with the rest of the slot. This notch
goes into the 1x portion of the 16x
slot. The characteristic notch makes
it easy to tell the difference between
PCI-E (PCI Express) and an AGP
Video card. As you can imagine, a
PCI Express Video card will not fit
into an AGP card slot, and a AGP
Video card will not fit into a 16x PCI
Express card slot.

Another physical difference between

PCI Express Cards, AGP, and PCI
is the distance between the card's
bracket and the start of the
connector. On PCI Express cards,
there is very little distance between
the metal bracket and the start of
the connector. On both PCI and
AGP the distance is much longer.

PCI Express 1x / 4x cards also have

the physical difference in regard to
bracket distance. They are both a
good deal smaller than standard
PCI. At this time this type of card is
still quite rare so at this point there is
little chance of confusing them with
something else. The PCI Express 1x
connector does however bare some
minor resemblance to an AMR slot,
so it is important not to confuse the
two. No motherboard on the market
today currently has both the older
AMR slot and PCI Express.

Q&A Common Questions about

PCI Express
Q:Is PCI Express Faster Than
PCI?
A:PCI Express is much faster than
PCI. For 1x Cards it at least 118%
faster. When you compare PCI
Express video to PCI Video the
difference is enormous: PCI Express
16x video is over 29x faster than

10
PCI Video.

Q:Is PCI Express Video Faster than AGP Video?

A:Yes and No. A 16x PCI Express connection is at least 190% Faster than AGP 8x but this is
the connection between the system and the video card. You use the connection the most
when your video card is low on memory or when the game you are using uses a Direct X or
Open GL feature that isn't supported in hardware.

So, what this means is that in terms of real world performance there may not be a huge
difference between AGP and PCI Express if you are talking about identical chipsets.
Unfortunately this is very hard to prove because graphics chipsets are designed either for PCI
Express or AGP. If you have a card that is available in both forms then you have a graphics
chipset that was designed for PCI Express and has a special bridge chip installed to let it
communicate with the AGP bus. The short of this is: if two cards of the same chipset are
available in AGP and PCI-E then the PCI-E one will always be faster. On PCI-E you don't
have the overhead of the bridge chip so it's faster, and you have the better bandwidth so in
intense situations such as high resolution gaming you'll come out on top every time.

The main point here is: If you have a system with AGP on it, it doesn't make sense to upgrade
just to get PCI-E video right now. The fastest AGP card to ever come out is likely to be the
nVidia 6800GT. If you are at a point where that is too slow then by all means it makes sense
to make a complete switch. If your happy with you're AGP graphics options, wait until you are
ready to upgrade the processor or other components before making the PCI-E switch. For
more information on AGP and PCI please see the general FSB guide.

Q:What is SLI?
A:SLI or Scalable Link Interface is a technology that lets you take two identical nVidia based
graphics cards *that support SLI* and a motherboard *that supports SLI* to achieve a very
high level of video performance. SLI works by splitting the rendering of the screen between
the two cards- one card renders half, the other card renders the other half. This technique is
extremely effective. For instance two 6600GT cards in SLI can do vastly better than a
6800GT or X800 card even though the price is lower for two 6600GT Cards. The downside to
this is that SLI is still new and is limited to systems based on AMD 64 / AMD FX Socket 939
processors. Needless to say though, there are some good solutions like this one from Asus.

Q:Do I need a special power supply for PCI-E [PCI Express].

A:Yes and no. Although the PCI-E spec calls for a PCI Express power connector, most PCI-E
cards don't currently use it. This means that you should only probably worry about this if you
are buying bleeding edge PCI Express parts. Card based on the ATI X600, ATI X700, ATI
X300, ATI X1300, Nvidia 6600, Nvidia 7600 or Nvidia 7300 series graphic chipsets rarely use
the connector. If you are in a situation where you need a PCI Express power connector but
the power supply doesn't have one you can always just use a PCI Express Power Adapter
that converts a 4-pin molex connector to PCI-E Power.

11