A New Standard

First released in 1992, the Peripheral Component Interface (PCI) has rapidly evolved into a viable replacement for the ISA bus, and is the most commonly used method for adding boards to a PC. It solves many of the problems with older architectures, while at the same time delivering a substantial increase in processing speed. PCI provides a new way of connecting peripherals to both the system memory and the CPU, with the goal of alleviating many problems encountered when installing new cards in an ISA based system (IRQ conflicts, address conflicts, etc.). To ensure the longevity of the bus, not only are systems based on newer PCI specifications backward compatible with those designed to older ones, PCI boards may also be used in a system that also uses ISA bus cards--however, this is becoming increasingly rate as ISA is rapidly becoming obsolete.

Strict Rules

To ensure that every PCI card will function properly in every PCI enabled system, a committee, the PCI Special Interest Group, was formed to set and manage standards for the bus. Quatech is a voting member of this committee, and all Quatech PCI products strictly adhere to the standards.

The most important of these standards is the requirement that all PCI cards implement specific configuration registers. Whether the PCI device is embedded on the PCI bus, or is an add-in board as Quatech's products are, it must include a unique "Vendor ID" and "Device ID," and a resource requirement list on its configuration registers. Any card which does not contain these registers cannot be considered a true "PCI" adapter, and might not work properly in your system. See Quatech's Guide to Selecting A Quality PCI Board for a detailed discussion of PCI compliance issues and a comparison of Quatech boards vs. non-compliant PCI boards.

Plug & Play

Implementing PCI control registers is vitally important to ensuring that Plug and Play, one of PCI's most attractive features, works properly. Setting jumpers and switches to configure address and IRQ is not required. The system configures itself by having the PCI BIOS access configuration registers on each add-in board at boot-up time. As these configuration registers tell the system what resources they need, (I/O space, memory space, interrupts, etc.), the system can allocate its resources accordingly, making sure that no two devices conflict.

Another implication of the PCI implementation is that a board's I/O address and interrupt are not fixed, meaning that they can change every time the system boots. Consequently, application software written for ISA boards, which are hardwired to particular interrupts, will not directly transfer to PCI-based systems. This is a serious consideration when contemplating switching to PCI. However, the alterations that need to be made to application software are usually minimal.

The High Speed, Wide Bandwidth Advantage

More than any other bus, PCI can take full advantage of today's high-power microprocessors to deliver extremely high speed data transfers. The original PCI bus was designed to operate with a 33MHz clock, to provide data transfer speeds up to 132 Mbytes/sec. These 32-bit adapters can use multiplexing to achieve 64-bit data transfers. (Later versions of PCI enable true 64-bit data transfers using up to a 133MHz clock to enable transfer speeds of up to 1066 Mbytes/sec.) These boards use a longer connector that adds an additional 32-bits of data signals. This is done by using the same set of pins to address and send data, the former implemented on the first clock cycle and the latter on the second. PCI's burst mode facilitates this operation as it allows a single address cycle to be followed by multiple data cycles. A special bus signal called a Cycle Frame is used to signal the beginning and end of a transfer cycle. Parity signals are used to ensure signal integrity, which is particularly vulnerable in such a complex transfer system.

The high speed data transfers across the PCI bus limits the number of PCI expansion slots that can be built into a single bus to 3 or 4, as opposed to the 6 or 7 available on the ISA bus. To expand the number of available expansion slots, PCI-to-PCI bridges are used. These bridges create a primary and a secondary PCI bus, each of which is electrically isolated from the other. Multiple bridges can be cascaded, theoretically allowing unlimited numbers of PCI slots to be configured in a single system. (Practically, it is important not to overload the CPU, as adding too many devices via expansion slots could considerably compromise system bandwidth.) The bridge enables bus transfers to be forwarded upstream or downstream, from one bus to another until the target/destination is reached.

Flexible Bus Mastering

Several pins on the PCI bus are reserved for implementing bus mastering. This means that any PCI device can take control of the bus at any time, even allowing it to shut out the CPU. Devices use bandwidth as available, and can potentially use all bandwidth in the system if no other demands are made for it. Bus mastering works by sending Request signals to the Central Resource (circuitry on the motherboard shared by all bus devices) when a device wants control of the bus, and when that control is ceded a Grant signal is received by the device. This flexible approach, which separates the arbitration and control signals (they were bussed together on MicroChannel and ISA), allows a computer designer greater control over the arbitration process.

Interrupt sharing on the PCI bus is also implemented to provide maximum flexibility. Four level-sensitive interrupts are located on the bus at pins A6, B7, A7 and B8, each of which can be assigned to from one to 16 separate devices. These interrupts can be activated at any time because they are not synchronized with the other signals on the bus. The PCI specification does not define how interrupts are to be shared. The process is implemented on a case-by-case basis by the motherboard manufacturer. For instance, Quatech multi-port serial PCI boards require only one slot to provide up to eight serial ports. These eight ports share a single interrupt, and our boards provide an interrupt status register that will indicate which of the eight ports triggered an interrupt.

PCI, UniversalPCI, and Voltage Requirements

The original PCI standard required that plug-in boards use 5V power provided by the PCs motherboard. As the PCI standard evolved, the option was added for a 3.3V power source. Now, with the latest PCI 2.3 release the 3.3V power supply is required and the 5V power supply is obsolete.

In order to make sure that a plug-in board receives the correct voltage (as getting the wrong voltage would cause the PC to be badly damaged) a different set of keys was designed for each voltage type. The key is the arrangement of gold fingers at the bottom of the board. Click to see a picture of PCI keying. When a board is inserted into a PCI slot, those goldfingers fit into the appropriate socket. Motherboards supplying 5.5V slots can only accept PCI cards keyed for 5V. Motherboards supplying 3.3V slots can only accept PCI cards keyed for 3.3V.

Both PC manufacturers and PCI board developers realize that users would optimally like a combination of both options. So, some motherboards provide universal connectors that can accept both 3.3V and 5V keyed PCI cards. Likewise, some PCI cards specifically designed to function with either a 3.3V or 5V power supply use a special key that will fit into either type of motherboard connector. These boards are called UniversalPCI boards.

Note that as stated above, the newest PCI 2.3 release has made the 5V power supply obsolete. This means that many of the newer computers can ONLY accept 3.3V PCI or UniversalPCI I/O cards. Choosing a UniversalPCI board, such as Quatech's serial UPCI line, provides added flexibility because it can be used in older and newer systems.

Low Profile PCI

Clearly, traditional PCI add-in boards are not practical for portable systems or for systems that use small-size cases. Two new standards have been incorporated into the PCI 2.3 specification that address these issues: Low Profile PCI and Mini-PCI. Low Profile PCI was designed to provide greater flexibility in desktop and server environments. The card is mechanically similar to a standard PCI card, but uses a shorter card and a different mounting bracket. (See Figure 3 below).

Figure 3: Relative sizes for PCI and Low Profile PCI boards

Low Profile PCI cards are designed to fit into systems as low as 3.350" with out using riser cards. There are two types of Low Profile PCI boards: MD1 and MD2. Both are built on 32-bit addressing, and differ only in length (MD1 being shorter than MD2.) Systems can be designed to support one or both configurations. Existing PCI backplanes used for standard cards can also accept the Low Profile cards. However, the Low Profile specification also includes a new bracket design that cannot be used with standard PCI boards.

Mini PCI, on the other hand, is not backward compatible with any older devices. It is designed to be used by system integrators to add additional functionality for mobile computers. It requires a completely new interface, and is typically used to add communication peripherals such as modems and network interface cards to notebook computers, docking stations, or sealed case PCs.

Mini PCI boards are even smaller than Low Profile boards, with a minimum size specification of 2.75" x 1.81" x 0.22." They are functionally equivalent to standard and Low Profile boards, using the same protocols, PC signals and software drivers. However, because of the small size they require higher density, more compact, and thus more expensive components. So, while Mini PCI is extremely useful for the mobile applications for which it was intended, it is not the most economical or flexible choice for desktop expansion.

PCI-X

The PCI-X specification is a high-performance enhancement to the conventional PCI bus specification. The first release of PCI-X double the maximum clock frequency that can be used by PCI devices from 66 MHz to 133Mhz, thus enabling communication at speeds over 1 Gbyte/sec. It also improved the efficiency of the PCI bus itself and the devices attached to it, by providing new features such as split transactions and transaction byte counts. PCI-X was developed for applications such as Gigabit Ethernet, Fiber Channel and other Enterprise server applications.

PCI-X 2.0, the current revision, increases speed and performance even further. Maximum clock frequencies of 266 Mhz and 533 MHz are now supported to permit lightning fast transfers of up to 4.3 Gigabytes per second of bandwidth--that is 32 times faster than the original conventional PCI 2.1 specification. PCI-X 2.0 also adds additional features for systems reliability that will minimize errors at high speeds and keep up with other advances in high-bandwidth architecture such as RAID, Fiber Channel, InfiniBand{tm}, and iSCSI.

PCI for Data Communication

PCI is the de facto standard for board-level expansion slots in PC-based systems. However, for serial and parallel communication, the potential of PCI goes largely untapped. Limitations of these communication protocols considerably slow down the system. The PCI specification anticipates this situation, and the bus is designed to slow down when dealing with low speed devices. So, though running serial and parallel devices via a PCI bus will not cause system problems, neither will it significantly improve system performance over ISA. But, as ISA slots are now essentially obsolete, you typically have no choice but to use PCI add-in boards in newer systems. For higher speed devices such as audio, streaming video, interactive gaming, high-speed modems, etc., PCI provides a clear advantage over older bus architectures.

In addition, PCs are becoming smaller and smaller in size, and many will not accommodate a standard PCI board, to say nothing of a huge old ISA board. For these systems, most notably Thin Clients and embedded systems, the Low Profile PCI form factor provides robust functionality on a very small size board.

PCI competes with USB for high-bandwidth applications. Both provide the advantage of Plug and Play installation, which may be important to some users. Currently, only USB supports hot swapping, but support for it is planned in a future PCI specification. USB has been positioned as a low cost solution for a variety of desktop applications. However, PCI still has a speed advantage even over the newer USB 2.0 when using true 64-bit data transfers. In fact, in many systems' USB ports are added to via the PCI Bus.

PCI Specs
Bus Clock Signal		133 MHz
Bus Width		64-bit
Theoretical Max. Transfer Rate		1 Gbytes/sec (8 Gbits/sec)
Advantages		very high speed, Plug & Play, dominant board-level bus
Disadvantages		incompatible with older systems, can cost more

Click here to see how PCI compares with other busses.