PCI Express 101

PCI Express-based workstations and graphics cards promise new levels of powerand performance, but there are a number of points to consider before jumping in.

Computer systems and their subsystems (such as graphics cards) need some wayto send and receive data. Since 1997 the conventional interconnect system forgraphics adapters in PC workstation systems has been the AGP (advanced graphicsport) standard in 1x, 2x, 4x, and 8x configurations.
These interconnects offer a maximum data bandwidth of slightly more than 2GBps,but they can only convey data in one direction at a time; that is, from the hostcomputer to the graphics subsystem or vice versa. But this can be a significantlimitation. For example, in order to maintain real-time frame rates, applicationsthat merge real-world video with computer-generated imagery require the abilityto transmit huge quantities of data in both directions simultaneously. Anotherissue with the conventional AGP bus standard is that it’s becoming increasinglydifficult to route these multibit parallel buses at the circuit board level whilekeeping all of the tracks one length and impedance.

There is a solution, however, and that is to move to a serial data communicationstechnology like PCI Express. A single PCI Express lane uses a set of differentialpairs to provide point-to-point connections between two devices: one to transmitand the other to receive data (Figure 1, below).

Figure 1: A single PCI Express lane allows for transmitting and
receiving using a set of differential pairs.

This single lane would be referred to as a x1 (“by-one”) implementation. Sucha lane can support up to 0.25GBps of real data communications in both directionssimultaneously, which equates to a total bandwidth of 0.5GBps. The PCI Expressstandard allows multiple lanes to be grouped together to increase bandwidth: x1,x2, x4, x8, x12, x16, and x32 lane widths are supported by the standard (x32 isnot currently available). A x1 connection might be used for a networking application;a x4 by a standard broadcast video capture card; a x8 by a high-definition videocapture card or a low-end graphics subsystem; while a x16—with a total bandwidthof 8GBps for 4GBps moving in both directions simultaneously—is required to satisfythe extreme bandwidth requirements of high-end 3D graphics subsystems.
Beware the “Gotchas”

Figure 2: A high-end PCI Express power connector for the additional needs ofa high-end graphics subsystem.

The high data bandwidths that can be achieved by PCI Express-enabled computersand graphics subsystems offer the potential for huge productivity gains in graphics-intensiveapplications such as CAD and digital content creation. However, there are manypotential “gotchas” that you must be aware of before purchasing.
When you purchase a computer that says it is PCI Express-enabled, what exactlydoes this mean? Different motherboards and supporting chipsets offer differentcapabilities. For example, take three PCI Express motherboards currently on themarket, one has a x16 connector and a x4 connector; the second has a x16 connectorand a x1 connector; and a third has a x16 connector and three x1 connectors. Additionally,some PCI Express motherboards may not even have a x16 connector, making do withtwo x8 connectors.
Thus, it’s important to ensure that the computer you purchase supports your specificPCI Express requirements and the physical size constraints for length, width,and height for the add-in cards you plan to use.
Similarly, when you purchase a new graphics card, it’s not enough to simply selectany card that says it’s PCI Express-enabled. Much like trying to squeeze the outputfrom a fire hose through a drinking straw, some graphics cards simply cannot takefull advantage of the bandwidth offered by PCI Express. In fact, it’s possibleto purchase a high-end AGP 8x-based card that significantly outperforms a low-to mid-range PCI Express-based offering.
And last but not least, there’s the fact that the PCI Express specification requiresthe motherboard to supply only 75W of power. This specification satisfies therequirements only of low-end and medium-range graphics subsystems. In order toaddress the needs of high-end graphics subsystems, the specification also allowsfor an additional (optional) 75W connector on the main computer’s power supplyto connect to the graphics card (Figure 2). The key word here is “optional,” whichmeans that you have to ensure the system can handle the power needs of your graphicscard before you bring them together.
PCI Express-enabled computers and graphics subsystems offer the promise of tremendousperformance and productivity gains. However, you first need to research the varioussubsystems to ensure that they are compatible, and especially that the graphicscard can take full advantage of the bandwidth provided by PCI Express technology.
Application Benefits
What are the true benefits to CAD end users? With the speed and data bandwidththat PCI Express provides, CAD professionals now have the ability to transmitdata to and from the graphics accelerator and process extremely large amountsof textures in real-time while manipulating their designs in a 3D environment.By harnessing the power of PCI Express, quick data transfer rates mean workflowswill not be interrupted with lagging display redraws.
Additionally, PCI Express helps accelerate CAD applications that render real-timeshader programs to photorealistic quality and mirror the material properties assignedto pieces of the design. These new photorealistic material shaders include glass,metal, water, wood, plastics, cloths, and other organic items. This industrial-strengthshader support, which is made possible by PCI Express’ fast data transfer rate,ensures that what is represented on your display is an accurate rendition of whatyou would see in the physical world.

Figure 3: PCI Express helps accelerate CAD applications that render
real-time shader programs to photorealistic images. It augments the 3D graphicspipeline.

It was also impossible to simulate fluid flow and dynamic calculations on a singleAGP or PCI in the past. PCI Express delivers the large bandwidth needed to performthese calculations completely on a fully programmable graphics accelerator (Figure3, above). This means faster completion of these calculations and the abilityto execute real-time flow analysis within your design.
Current CAD software applications are written so that heavy processing and computingtasks are sent to the CPU first. This means the CPU calculates and compiles data,then sends results to the graphics card for additional manipulation or final rendering.While the CPU is churning, the graphics card is idle. Today’s graphics cards includemore onboard memory reserves and faster visual processing units (VPU) that computegraphics-intensive data faster than a CPU.
PCI Express can be a huge help in this situation due to its large data transfercapabilities. As software developers design new applications that use the graphicsaccelerator to process data and then push the resulting data to the CPU for furtherprocessing, more bandwidth between the system and graphics accelerator will berequired. Thanks to the improved bandwidth offered by PCI Express, design professionalswill see a huge increase in overall productivity.
Computer hardware is in a constant state of evolution. With PCI Express, savvycomputer shoppers should be able to choose graphics, motherboard, and processingsolutions that give them more room to grow.
Clifton Robin is the brand manager at 3Dlabs. He has more than 19 years of experience in hightech. You can contact him via e-mail sent to [email protected].

Product Information
3Dlabs
Milpitas, CA
3dlabs.com