Jumping off the PCI Bus

OPINION: When new PC standards come online there is a danger to assume that the benefit will come from an immediate performance boost. Danger because one of the defining features that sets modern standards apart from legacy ones are that they are now being developed with scalability in mind.

Technologies like Parallel ATA and PCI were never intended to last as long as they did. While PCI just passed its tenth birthday, Parallel ATA first appeared 20 years ago as an option on IBM's PC-AT.

Both of these technologies have been scaled far beyond their initial capacities and now Parallel ATA is in the process of being superseded by Serial ATA and PCI is about to be usurped by PCI-Express.

Over the past year, Serial ATA has become more and more common; however it has to fight against the fact that it offers no measurable performance increase over existing Parallel ATA hard drives.

While SATA has a theoretical bandwidth of 150MB/s, the current physical limitations involved in hard drive construction mean that actual data transfers are well below the 100MB/s offered by Parallel ATA.

However, SATA offers other advantages that are less easily shoehorned into a graph, like much improved cabling, more reliable signalling and greater ease of use. SATA is slowly overcoming these perception problems, but it will be a while before we see the bandwidth being taxed.

Sometime during the first half of this year we are expecting to see PCI-Express finally move out of the testing labs and into our PCs. This has been a hugely hyped move, but there are a lot of misconceptions flowing around about what PCI-Express is and what it means for end users.

One of the best ways to conceptualise the difference between PCI and PCI Express is to think of networking hubs and switches. When running via a hub, network traffic is much less efficient, packets bounce off each other and everything bottlenecks at the hub. This happens on the PCI bus, and it limits the overall amount of traffic that can be run over it.

To combat these problems with PCI, a few years ago Intel created a new AGP bus designed to give video cards a sped up, dedicated laneway to the rest of the PC, bypassing the bottlenecks of PCI. But even AGP bandwidth is woefully insufficient for the potential of today's video hardware, and more and more data hungry devices like Gigabit Ethernet are vying for the PCI bus' attention.

The solution comes in the form of PCI-Express, which throws away the hub and replaces it with a switch. This allows for smooth flow of traffic at maximum speeds, even without adding anything above and beyond the switching hardware. But PCI-Express goes much further than that by introducing a 'lane' structure for data.

This is done in acknowledgement of the fact that the bandwidth requirements for various types of hardware differ dramatically. In fact, PCI-Express will launch in the form of two connectors, a single lane (x1) connector for Gigabit Ethernet and a 16 lane (x16) connection that will replace AGP as the video card connection of choice.

If we take Gigabit Ethernet as an example, under PCI it can only ever achieve a transfer rate of around 900Mb/s. However, when Gb Ethernet is moved over to a 1x PCI-Express connection then it is capable of throughput in the range of 2Gb/s.

Over time more and more hardware will move onto the PCI-Express bus, but for now it only really provides bandwidth advantages for a small range of devices, like graphics and Gigabit Ethernet. But it is scaleable and that is what really counts.

While current operating systems will not see any difference between cards connecting via PCI-Express and PCI, there are further technologies built into the PCI-Express specifications that will be tapped in the future by the next generation of Windows, codenamed Longhorn.

So is the way with the new generation of PC standards? After the ridiculousness experienced with the introduction of USB (and USB 2.0) to the market eyes are now open to the fact that waiting for software support to materialise is not the way to go.

This rapid introduction of new technology, followed by a gradual introduction of support for the new features means that the end user can happily jump onto a new technology now without purchasing a new operating system or going through some sort of driver installing contortionism. It is the way things run nowadays, and has generally made the life of the early adopter much easier.

PCI-Express is on track to launch sometime in the first half of this year on Intel's next set of desktop chipsets, codenamed Tumwater and Grantsdale. Accompanying this will be a new generation of video cards from the big players ATI and NVIDIA, and hopefully affordable Gigabit Ethernet from a variety of vendors. It'll be a slow introduction, but a much needed one.