How does PCI Express work

PCIe - PCI Express (1.1 / 2.0 / 3.0 / 4.0 / 5.0)

PCI Express (PCIe) is a fast internal interface for expansion cards in computer systems. With the introduction of PCIe in 2004, the AGP as a graphics card interface was put to an end and PCI was also replaced as an internal computer bus system.

In addition to being used in desktop computers, there is a variant, PCI Express Advanced Switching, which is used as a backplane in modular servers. PCI Express is basically suitable for connecting several servers and assemblies to form a computer system or telecommunications server. This technology is suitable for time-critical applications such as image and sound reproduction. Functions for reserving minimum bandwidths are also available.

PCI Express is intended for copper lines and optical connections. The PCI Express specification describes the software protocol, electrical and mechanical properties of the connectors and expansion cards. Some of this is known from the serial connection systems such as USB and FireWire. For example, plugging and unplugging during operation (hot-plug) and bundling several cables to increase the transmission rate.

PCI Express will become even more important in the future. The semiconductor circuits from PCI Express are also used in a modified form for DisplayPort, SATA and SAS. In the future, SATA will be replaced by an SSD interface called SATA Express, which is significantly influenced by PCI Express.

Slot / slots

There are different slots or slots for PCIe on the motherboard. The long x16 slots for graphics cards and the short x1 slots for different expansion cards are typical. Some motherboards also have x4 and x8 slots.
The designation x1, x4, x8 and x16 indicates how many PCIe lanes are cascaded in the slot.

Basically, short cards also work in long slots. But be careful, the mechanical length says nothing about how many lanes a slot has. A x16 slot can only have 4 (x4) or 8 (x8) lanes. On some boards, several slots share the lanes.

Architecture or topology of PCIe

Classic bridge model
(Bus topology)
Modern switch model
(Star topology)

In contrast to PCI, with the bus structure in which all connected components have to share the available bandwidth, with PCI Express serial connections are switched to a switch that is located in the chipset. The switch connects a PCIe module directly with the main memory or other modules with full bandwidth and speed.
At the logical level, PCIe is fully compatible with the old PCI. The operating system does not notice any difference. Not much of PCIe can be seen in the Windows Device Manager either.

Instead of parallel bus systems, serial point-to-point connections are used. The old bus topology was already replaced by a star topology with Ethernet. A central switching point (switch) connects two devices directly with each other.

The reasons for switching from the tried and tested bus structure to serial point-to-point connections are the huge number of address and signal lines. The increasing number of signal lines on the motherboard requires a lot of space, combined with high power consumption. The transmission speed cannot be increased arbitrarily either, because the parallel lines influence each other (crosstalk).

Transmission technology

The transmission technology of PCI Express is based on two differential line pairs (4 wires), which are referred to as a link or lane. One pair of lines for sending data, the other for receiving data. To increase the speed, a device can use several lanes. A total of up to 32 lanes can be bundled. In practice, however, it appears that simple expansion cards only have one lane. Exceptions are graphics card slots called PEG (PCI Express for Graphics). You have 16 lanes available.
The parallelization of the data does not take place on the electrical, but on a higher protocol level. Differences in runtime, line disruptions and failures are also compensated here.

Transmission speed

The transmission speed with PCIe is "oriented" to the version and the number of links or lanes. The higher the version and the more lanes, the higher the bandwidth and the higher the transmission speed.
The bandwidth indicates how much capacity is theoretically or maximally available for data transmission. However, the actual data rate is lower.

PCIeSymbol rate
per lane
PCIe x1PCIe x4PCIe x8PCIe x16Coding
(Weight in%)
1.0/1.12.5 GT / s0.25 GB / s1.0 GB / s2.0 GB / s4.0 GB / s8b10b / 20%
2.0/2.15 GT / s0.50 GB / s2.0 GB / s4.0 GB / s8.0 GB / s8b10b / 20%
3.0/3.18 GT / s0.97 GB / s3.9 GB / s7.8 GB / s15.5 GB / s128b / 130b / <2%
4.016 GT / s1.90 GB / s7.8 GB / s15.5 GB / s31.5 GB / s128b / 130b / <2%
5.032 GT / s3.90 GB / s15.5 GB / s31.5 GB / s63.0 GB / s128b / 130b / <2%
6.064 GT / s7.50 GB / s30.1 GB / s60.2 GB / s120.4 GB / sPAM-4

Explanation of the specified speed information: The information on the transfer rates is rounded. If one assumes a symbol rate of 5 GT / s (gigatransfers per second) with PCIe 2.0, which corresponds to a bandwidth of 5 GBit / s per lane, then the bandwidth is reduced by the 8B / 10B coding (10 bits per byte ) to about 4 GBit / s. This corresponds to a net bandwidth of 0.5 GByte / s per direction. The actual data rate is then even lower. Because in addition to the pure data transmission, a transmission protocol with commands, addressing and confirmations is active that uses part of the bandwidth, which is why the actual data rate is once again below the net bandwidth.

PCI Express graphics card compatibility

PCIe is fully up and down compatible. This means that old cards fit into new motherboards and vice versa. An x1 card also works in an x16 slot and vice versa. Provided that it fits mechanically.
To ensure compatibility, the PCIe host controller negotiates the number of lanes and their transfer rate when the expansion cards are initialized.

Card / slotPCIe x1PCIe x4PCIe x8PCIe x16
PCI Express x4-OK??
PCI Express x8--OK?
PCI Express x16---OK

Ok = compatible / - = not compatible /? = not mandatory but possible

Note: Basically, the PCIe standard provides for downward compatibility. But that doesn't mean that all products adhere to it.

PCI Express 1.0 / 1.1 (PCIe 1.0)

There is practically no difference between PCIe-1.0 and PCIe-1.1. The speed per lane is set at 2.5 GBit / s, which corresponds to a net bandwidth of 250 MByte / s.

PCI Express 2.0 / 2.1 (PCIe 2.0)

The speed has been increased to 5 GBit / s per lane. In the best case, a net transfer rate of 500 MByte / s can be achieved.
PCIe 2.0 is backwards compatible with PCIe 1.0 / 1.1. But be careful, PCIe 1.1 cards should work in PCIe 2.0 slots. If there are problems, a BIOS update may help. There is a special feature in the specification of PCIe 2.0. There it is stipulated that PCIe 2.0 cards must also work in PCIe 1.1 slots. There is no structural difference.
The main difference between PCIe 2.0 and the older version is the maximum possible transfer rate. PCIe 1.1 works at 2.5 Gbit / s per lane.

PCI Express 3.0 / 3.1 (PCIe 3.0)

Thanks to the reduction in the overhead for data transmission and a more efficient line code, the transfer rate could again be doubled compared to PCIe 2.0. This means that each PCIe lane has a bandwidth of 8 GBit / s and transmits 1 GByte / s.
There aren't many applications that push PCIe 3.0 to the max. Except for graphics cards, PCIe 3.0 only makes sense for a few expansion cards. Most likely for 40 Gigabit Ethernet cards or host adapters for server mass storage. With PCIe 3.0 graphics cards can communicate with the chipset or processor over 16 lanes at up to 16 GByte / s.
However, the speed of PCIe 2.0 can be short if the PCIe is the central bus system between the processor and the chipset. If the connection between CPU and chipset only supports 2 GByte / s in both directions (PCIe-2.0-x4), PCIe 3.0 is only available for external graphics cards, which are then connected directly to the main processor. In highly integrated processors from Intel and AMD, the PCIe Root Complex ends or begins in the processor. There is no getting around PCIe 3.0 for the future. In addition, the PCIe will replace SATA as a mass storage interface. The PCIe lane concept is clearly superior to SATA. The PCIe lanes can be scaled more easily.

PCI Express 4.0 (PCIe 4.0)

The PCI Special Interest Group (PCI-SIG) completed the PCIe 4.0 specification in 2017. Compared to PCIe 3.0, the data transfer rate doubles to 16 GBit / s and allows a maximum net data rate of around 2 GByte / s per lane. PCIe 4.0 x16 then manages almost 32 GB / s.

The need for faster PCIe connections is not determined by the CPU and GPU manufacturers, but by the storage and network industry. This is where PCIe 3.0 comes to the limit with 16 parallel PCIe lanes for network cards beyond the 100 GBit class. And NVMe SSDs in the enterprise sector are limited to four parallel PCIe lanes due to their m.2 or u.2 connection. More bandwidth is only possible if you increase the transfer rate per lane.

To enable this transfer rate, the maximum cable length is reduced from 20 to 8 to 12 inches (20 to 30 cm). New materials for conductor tracks and contacts are also required in order to maintain the signal quality for this speed.
PCIe 4.0 only achieves its maximum data rate if a maximum of one plug connection hangs in the line path (point-to-point connection).

As expected, PCIe 4.0 should continue to be downward compatible with older PCIe cards. Normal PC users will rarely have to deal with PCIe 4.0 devices because the higher speed is more interesting for the professional sector.

PCI Express 5.0 (PCIe 5.0)

Because the need for more speed will continue to increase in the future, the PCI-SIG for PCIe 5.0 has the prospect of another doubling of the bandwidth to around 32 GBit / s and thus around 4 GByte / s per PCIe lane.
Presumably this can only be achieved if the chips involved are soldered on the same board.

PCIe 6.0

In order to double the data rate again to 64 GT / s, PCIe 6.0 provides for the use of pulse amplitude modulation (PAM-4) and the implementation of predictive error correction (RS-FEC). As a result, the net bandwidth is not exactly doubled, but it is only slightly lower.

Overview: internal bus systems

Other related topics:

Everything you need to know about computer technology.

Computer technology primer

The computer technology primer is a book about the basics of computer technology, processor technology, semiconductor memory, interfaces, data storage devices, drives and important hardware components.

I want that!

Everything you need to know about computer technology.

Computer technology primer

The computer technology primer is a book about the basics of computer technology, processor technology, semiconductor memory, interfaces, data storage devices, drives and important hardware components.

I want that!