PCI. PCI-X 1.0. PCI-X 2.0. PCMCIA. PCI Express. Do you know the difference?
The truth is, most of us don't, despite the fact that the local bus is one of the most important parts of modern computing, crucial to the speed of client/server systems. And worse, nearly all of us confuse PCI-X, short for PCI Extended, with PCI Express, the newest implementation of the PCI bus and a true revolution in the way data is moved from peripheral to processor.
In The Beginning Was PCI
To understand PCI-X and PCI Express, you need to understand plain old PCI, or Peripheral Component Interconnect. It's the dominant local bus used in servers, clients, communications, and embedded applications, not to mention newer versions of the Macintosh. And it's the base that PCI-X and PCI Express are built on.
Developed by Intel in 1992 to improve on the ISA bus, PCI is a 64-bit technology that's usually implemented as 32-bit. It can run at clock speeds of 33MHz or 66MHz. At 33MHz, 32 bits translates into a throughput of 133Mbps.
And while PCI-X and PCI Express are both improvements on PCI, don't expect PCI to go anywhere anytime soon. In much the same way that PCI and ISA slots can often be found side by side on the same machine, neither PCI-X nor PCI Express is designed to replace PCI.
Often confused for PCI Express, PCI-X stands for PCI Extended, a standard designed jointly by HP, IBM, and Compaq to increase the performance of high-bandwidth devices, such as Gigabit Ethernet and Fibre Channel, as well as processors that are part of a cluster. It's useful to think of it as an enhanced form of the PCI bus.
PCI-X is fully backward compatible with PCI and improves on the throughput of PCI from 133Mbps to 1Gbps. It comes in two flavors: PCI-X 1.0, with a maximum clock speed of 133MHz, and PCI-X 2.0, which tops out at 533MHz. (In fact, PCI-X 2.0 has four "speed grades": PCI-X 66, PCI-X 133, PCI-X 266, and PCI-X 533, which run at 66MHz, 133MHz, 266MHz, and 533MHz, respectively; which grade is used depends on the application of the bus.)
But 2.0 also has other refinements beyond speed, including ECC (Error Correction Code), to improve the robustness of the interface; 1.5V signaling and source synchronous strobes, which improve performance so it can run at 533MHz; device ID messages, which are designed to enable a new class of peer-to-peer transfer applications; and a 16-bit version for embedded applications.
Servers and workstations are the initial targets for PCI-X 2.0. InfiniBand, 10 Gigabit Ethernet, 10 Gigabit Fibre Channel, and multifunction cards that have large bandwidth requirements will benefit from the PCI-X 266 and PCI-X 533 flavors. The PCI-SIG, the ultimate authority on the use and deployment of the technology, is mum on whether PCI-X 2.0 will make it to the desktop segment of the market. That, it says, will be determined by the needs of individual developers and their products.
In the past few years, a slew of new technologies, such as 10GHz processors, faster types of RAM, high-performance graphics, multiple-Gbps networking, fast storage, and more have begun to demand a faster interconnect, one with greater bandwidth to service their hungry needs. Enter PCI Express, a high speed, serial, point-to-point, hot-pluggable, and hot-swappable system bus that's fully compatible with PCI, at least on the software level.
Each point-to-point interconnect in the PCI Express architecture can have 1, 2, 4, 8, 12, 16, or 32 dual simplex 2.5Gbps lanes (with a 2.0Gbps effective rate), which translates into scalable bandwidth up to 128Gbps (or 16GBps) between nodes.
PCI Express is also called 3GIO (Third Generation I/O) and occasionally Arapahoe, after the IEEE working group where it was first developed as a joint project with Intel at the head. In early 2002, ownership of the technology was transferred to the Portland-based PCI-SIG where it was renamed PCI Express Architecture. Products with PCI Express are expected sometime in the early part of this year (with initial silicon in the second half of 2003).
PCI Express is not intended to replace PCI or PCI-X. Rather, the PCI Special Interest Group will continue to support both standards because both have attributes that are needed on a per-application basis. As proof of this, consider the fact that PCI/PCI-X and PCI Express slots will coexist in the same machines in the future and be used for different types of interconnects. (PCI Express is, however, intended to replace AGP.)
Some key features of PCI Express include software compatibility with PCI, meaning a smooth transition to new hardware can be made with software that leverages the new, advanced features of PCI Express, and optimum bandwidth per pin, to enable unique and small form factors, simplify the design of motherboards and the way they route signals, and reduce problems with signal integrity. There's also native hot-plug/ hot-swap capacity and native power management capacity.
All in all it's a powerful and promising technology, the exact adoption of which remains to be seen. But along with PCI-X, it's sure to make server rooms and data centers faster the world over.
by David Garrett