2004 - PCI Express: Interconnection For the Next Decades
The start of transition to PCI Express bus from PCI in 2004 was generally a regular event as the industry knew it was coming. However, the significance of the bus, prospects that it enabled for the industry as well as changes that its deployment caused are hard to underestimate.
The original PCI (peripheral serial interconnect) bus introduced in 1993 was a revolution for its time. While PCI bus is still used by many devices, it became obvious even by 1997 (which is when AGP was introduced) that its bandwidth is not enough for graphics accelerators, moreover scalability of PCI's bandwidth was pretty limited and it could not grow as fast as demands from various peripherals. While PCI-X solved the problem for servers, it was a rare guest on desktop computers and was could not be used on notebooks. Finally, parallel approach to data transfer, where a large number of slow-speed links is required meant that add-on cards were relatively large because of large connector. So, sometime in 1999 engineers from Intel as well as Compaq (now HP), IBM and Dell started to design a new serial high-performance data interconnect bus code-named Arapahoe.
The PCI Express bus is based on point-to-point serial links and hence it is very scalable both in terms the amount of lanes as well as clock-speed. It features a number of innovations not available previously on any internal buses. Nonetheless, the PCIe fully maintains software compatibility with conventional PCI. Already the first version of the standard provided 250MB/s of bandwidth per single PCI Express lane, compared to 33MB/s provided by PCI 32-bit/33MHz.
The increased bandwidth and ability to configure and reconfigure the number of lanes dynamically not only increased performance of personal computers, but also opened the door to a number of technologies and devices previously impossible on the desktop.
The PCI Express bus allowed graphics chip designers to return multi-GPU setups to the market, something impossible with AGP. Makers of advanced solid-state drives and high-end RAID controllers are no longer bandwidth-limited. It is also now possible to create 40Gb Ethernet or 100Gb Ethernet cards. There are a number of other bandwidth-hungry applications that benefit greatly from PCI Express version 1.x and 2.x and the version 3.0 is projected to bring a number of protocol-related improvements, which will particularly help to improve efficiency of GPGPU-based applications.
But not all companies could benefit from the arrival of PCI Express. For example, Creative Technology, once the driving force of high-quality audio on the PC, failed to deliver audio processors that would benefit from PCIe innovations on time assuming that extra bandwidth was not needed for audio. As a result, at some point the company discovered that its expensive audio cards could not compete against inexpensive integrated solutions as their advantages were not obvious to the majority of end-users. Obviously, PCIe bus alone could not help Creative much, but at least it would have paved the way for some creative thinking within the company and could have resulted in new features. A number of other significant names of the past also could not benefit from PCI Express revolution and failed into oblivion.
Starting 2011, some chipsets from Intel Corp. will lose support for PCI, which will mean end-of-life for the bus after 17 years of existence (11 of which it was the primary bus for PCs). By contrast, PCI Express technology is being actively developed even on its sixth year on the market. As a result, it may eventually outlive its predecessor significantly.