At least that’s what marketing departments would have you believe. In reality, there’s a reason why folks who jump on recently released gadgets are called “early adopters.” For the same reason that knowledgeable auto aficionados often hold off on the first model year of an anticipated car, computer hardware enthusiasts recognize that the initial iteration of any technology will have its idiosyncrasies and teething pains. The PCI Express bus topology, pioneered by Intel, is no exception. High-end motherboards abound, but good luck locating a comparable graphics card for customers who’d be interested in upgrading.

Fortunately, the transitional period between PCI Express’s introduction and its whole-hearted acceptance looks to be relatively short. Compatible chipsets and mainstream graphics cards began emerging several months ago, Intel’s E7525 workstation chipset is readily available, and now there’s a new breed of graphics cards hitting the market targeting the professional with heavy-duty demands.

Boldly Going Where No Reseller Has Gone Before

The same folks who were likely to buy an AGP adapter for their graphics workstation will be the ones interested in PCI Express. Jeff Little, director of marketing at 3Dlabs, points at ultra high-end graphics markets, such as automotive body styling, medical imaging, oil research, seismic imaging, and any number of digital content creation applications.

For the most part, all of those markets are already being addressed by professional AGP 8x adapters, though. Where does PCI Express fit in and how can you make it more compelling? Consider the benefits of PCI Express: bi-directional communication, high data rates for an improvement in overall bandwidth, and incredible scalability to support any number of devices in the same system, each with point-to-point access to the host switch.

According to 3DLabs’ Little, the one market most likely to benefit from PCI Express in the short term is broadcast video editing, specifically high-definition content. Current solutions are, for the most part, custom designs involving an AGP graphics card and a PCI HD card connected through an “over the top” ribbon cable or connector. There aren’t any standards defining how those components work together, and as a result it’s a highly inaccessible market for system builders, despite plenty of potential for margin. The bandwidth enabled by PCI Express eliminates the need for proprietary connectors and complicated configurations.

For example, a professional PCI Express graphics card in a x16 slot, together with an HD card populating a x4 slot, is able to transfer three uncompressed high-definition streams simultaneously. Little claims that’s the holy grail for HD video editing, and the hardware to make it happen is nearly completed. The workstation-class graphics cards are available today and PCI Express HD cards are expected in the next six to nine months. And that’s just one example of a highly specialized market being pulled within reach of small system builders through the intelligent implementation of new technology.

Of course, you probably won’t see a dramatic upswing in sales thanks to high-definition video editors piling out of the woodwork. So, how can you use PCI Express today to motivate your customers? Offer them more performance, of course.

3Dlabs: Back With A Vengeance

Of those manufacturers with a prominent role in professional graphics, 3DLabs has the most experience. And despite the far-reaching resources of both ATI and NVIDIA, 3DLabs continues to focus its energies exclusively on optimized workstation solutions – and succeeds it. The company’s latest efforts center on a redesigned graphics processor, codenamed P20 and officially dubbed the Wildcat Realizm VPU.

P20 is a significant improvement over the P10 that precedes it. Equipped with 12 pixel pipelines (versus P10’s four), the Wildcat Realizm would appear to have significantly more fill rate. However, its purpose is geometry processing, enabled by sixteen 36-bit, highly accurate vertex shading engines. The chip is manufactured on an advanced 130nm process and weighs in at 150 million transistors, rivaling ATI’s popular X800 with regard to sheer size. Like some of the gaming architectures that bridge over to professional use, 3DLabs’ Realizm supports DirectX 9.0 (Vertex Shader Model 2.0 and Pixel Shader Model 3.0), OpenGL 1.5, and driver support is planned for OpenGL 2.0. 3DLabs also claims that 64-bit drivers will be made available for P20 once the operating system infrastructure falls into place.

That’s all well and good of course, but none of the new chip’s specifications will necessarily stand out to resellers who’ve already seen NVIDIA’s Quadro FX or ATI’s high-end FireGL cards in action. What will set your customers’ hearts ablaze is the high-end implementation of P20, currently 3DLabs’s only PCI Express product, named the Wildcat Realizm 800.

According to 3DLabs, the bandwidth enabled by PCI Express makes it possible to put two Wildcat Realizm VPUs on a single card. That simply wasn’t possible with AGP 8x because the interface offered limited bandwidth and was only able to communicate in one direction at a time. If you thought it’d take PCI Express years to become useful, 3Dlabs’ Realizm 800 is evidence that the bus topology is already being deployed in full force.

Each of the Realizm 800’s visual processors interfaces with a 256-bit memory bus populated by 256MB of GDDR-3 memory, forming what 3DLabs likes to call a 512-bit bus. Both Wildcat Realizm VPUs, in turn, connect to 3DLabs’s VSU (Vertex Scalability Unit), which allows the natively AGP 8x processors to communicate over a PCI Express bus. The VSU also assumes geometry processing duties with a pair of vertex shading engines, each wielding 16 processors. Because the VSU runs independently of the VPUs, it has its own 128MB GDDR-3 memory repository (giving the board a total of 640MB), termed DirectBurst, for storing fragment shader instructions and command data.

Geometry information leaving the VSU travels over dual 64-bit buses to each VPU, which display the rendered output in a 64x64 checkerboard pattern. The processors render every other block in the checkerboard, effectively balancing the workload between themselves. The Realizm 800 outputs the final display to two dual-link DVI-I connectors.

The Wildcat Realizm 800 is a large card. Its oversized cooling solution occupies one PCI Express x16 slot and an adjoining expansion space. All of the onboard components require about 120W of power, necessitating an auxiliary input. And its $2,799 MSRP is a razor-sharp indication that 3Dlabs means business. Nevertheless, its size, power needs, and price are all relatively tame compared to NVIDIA’s dual-card alternative.

NVIDIA: Going Pro With
Gaming Tech

The past six months have been good to NVIDIA. The GeForce 6-series desktop architecture is a vast improvement over last generation’s best effort, and the migration to PCI Express seems to be going fairly well for both desktop and workstation products. In fact, NVIDIA timed its high-end graphics rollout fairly well to coincide with Intel’s E7525 chipset and its own AMD64 platforms.

NVIDIA leverages a broad range of graphics processors and a special AGP-to-PCI Express bridge chip to make the connection between older hardware and the latest chipsets. Consequentially, there are plenty of workstation-class PCI Express cards for any combination of price and performance. But like 3Dlabs, NVIDIA’s most exciting developments occur at the ultra high-end, where budgets take a backseat to your customers’ need for speed.

Rather than incorporate two of its processors onto a single graphics board, NVIDIA uses two separate cards to render cooperatively, a technique once employed by enthusiast giant 3dfx before going out of business. Having acquired 3dfx, NVIDIA now uses the same marketing name–SLI–to describe its newly updated dual-card implementation. The nuts and bolts are different. NVIDIA’s SLI divides the screen horizontally and runs a load-balancing algorithm to divide the workload evenly, while 3dfx had each card render alternating scan lines. However, the end result is the same. Both technologies yield incredible gains in applications limited by graphics performance.

The newest PCI Express Quadro FX cards all support SLI, including the 4400, 4400G, 4300, and 1400. Match identical models from the same manufacturer, find a motherboard with two PCI Express x16 graphics slots, connect the card’s with NVIDIA’s proprietary over-the-top SLI connector, and you’re good to go.

Of course, there are both advantages and disadvantages to using SLI. The technology is attractive because allows you to use one graphics card today, in much the same way that you can sell a dual-processor workstation to a customer with just one CPU installed. SLI enables scalability without the immediate financial commitment. And although a single Quadro FX 4400 doesn’t look to be as fast as 3DLabs’ Realizm 800 (based on benchmark numbers provided by both companies), two of those Quadro cards working together will outperform any competing implementation.

The appeal of SLI diminishes somewhat when you consider its requirements: a specialty motherboard with two graphics slots, extraneous power delivery, and the price of two expensive video cards. But for those who are solely interested in maximizing the muscle of their graphics subsystem or truly intend to add a second card at a later date, SLI is still a compelling idea.

In Retrospect

Although PCI Express graphics cards offer limited immediate benefits to the desktop market, there are some interesting new technologies on the professional side capable of multiplying performance. In order to utilize those ultra high-end products from 3Dlabs and NVIDIA, your customers will need a new motherboard, at the very least. Processors, memory, and power supplies are also likely upgrades you’ll want to suggest. PCI Express clearly paves the way for plenty of hardware sales, but will your customers be satisfied with their new investment? Those who are already pushing the limits of today’s professional cards will indeed see a marked improvement in rendering performance.