RAM | Reseller Advocate Magazine: Issue 61

By Chris Angelini

THE GRAPHICS INDUSTRY IS AT A BIT OF A crossroads. ATI and NVIDIA have been fighting for supremacy of the DirectX 9 market for more than four years now—a veritable eternity when you're talking video cards. ATI struck first with its Radeon 9700-series. NVDIA swung back with the GeForce FX-series, which initially missed its mark almost entirely. ATI focused attention on its killer next-gen architecture, stuttered, and allowed NVIDIA time to hit back with the GeForce 6-series. Ever since, the two have unveiled successively faster products, refreshing feature sets to reflect Microsoft's subtle addendums to DirectX 9 but otherwise systematically exchanging blows in a big build-up to the next best thing.

That next best thing began life as the Microsoft Windows Graphics Foundation, was changed to DirectX Next, and is now referred to as DirectX 10, a collection of multimedia-oriented APIs (application programming interfaces). Exclusive to Windows Vista, DirectX 10 effectively forces your customers to upgrade their operating systems in order to experience its special features at full speed. Those who resist the transition will be forced to stick with a combination of Windows XP and DirectX 9.Don't expect many to hold out for long, though. Vista is the first OS from Microsoft with a 3D-accelerated desktop, which looks to be pretty sweet.

NVIDIA and ATI/AMD are counting on the upgrade being worthwhile. After all, Vista and DirectX 10 introduce quite a bit of new functionality. Both graphics vendors are readying their freshman next-generation offerings able to utilize all DirectX 10 has to offer. The difference is that NVIDIA's solution is available to your customers today, in time for holiday shopping, and ATI's is still being finalized with an expected release date to coincide more or less with the consumer Windows Vista launch.

Why DirectX 10?

Today's DirectX 9 graphics processors are highly programmable, letting software developers explore the limits of their creativity in writing custom effects programs, often referred to as shaders. The logic needed to process those shader programs has historically been divided into two sections of the GPU—one for addressing geometry and another tasked with pixel processing.

Over time, a couple of things have happened, though. First, the hardware used to address pixel and vertex shading developed very similar capabilities through the evolution of Shader Model 2.0, 2.x, and 3.0. Second, it's becoming more common for certain scenes to incorporate very complex geometry or intensive pixel processing, leaving the opposing vertex or pixel shaders fundamentally idle.

An Enabling Platform
Building the right machine for Windows Vista and DirectX 10 requires the right motherboard and chipset. NVIDIA’s new nForce 680i SLI is perfect for enthusiasts eying Core 2 Duo processors and GeForce 8800-series cards.

DirectX 10 alleviates any potential for bottlenecking by unifying vertex and pixel shading. (Those two components were infamous for their redundant functionality.) So rather than having a vertex shader fully taxed while the pixel shaders sit still, one larger resource pool is responsible for processing both shaders, greatly improving performance as bottlenecks fade away. And because DirectX 10 has been completely rewritten to take advantage of a unified architecture, a lot of the processing overhead that found its way into DirectX 9 is also kept to a minimum, effectively increasing the number of unique objects that can appear on-screen at any given time.

Beyond the performance enhancements DirectX 10 promises to deliver, improved stability due to driver implementation should be another one of the points you mention when it comes to selling DirectX 10 hardware. Microsoft's new driver model is split between user-mode and kernel-mode, whereas DirectX 9 makes very heavy use of the operating system's kernel space. By putting as much Direct3D, OpenGL, and video playback code into the user-mode space as possible, Vista should be much more adept at recovering from crashes.

DirectX 10 Positioning

The move from DirectX 9 to DirectX 10 is significant both to the hardware and software industries. Fortunately, game developers already have experience writing for a unified architecture since the Xbox 360, powered by an ATI graphics processor, introduced the concept a little over a year ago. Consequently, the software folks, who normally have to wait for hardware before they can begin development, have a head start. Instead of trying to sell your customers hardware with a caveat that the games written for their $500 video cards are expected a couple of years down the road, they can expect to see DirectX 10 titles within months of Vista availability.

A Whole New World
The GeForce 8800 graphics processor is radically different from any other design NVIDIA has ever endeavored. Clusters of stream processors constitute the chip’s unified shader architecture and 64-bit memory pathways yield a 384-bit bus.

Now, there's a good chance you've heard graphics vendors positioning their DirectX 9 cards as "Vista Premium-ready" over the past year or so, convoluting the compatibility issue to a degree. Vista does in fact feature DirectX 9.L—a legacy mode—running alongside DirectX 10, allowing older graphics cards to render the most graphically intense Vista user interface. So, all of those mainstream Radeon X1600- and GeForce 7600-series cards you'll continue selling until less expensive DirectX 10 boards emerge are wholly ready for Vista when it drops. They simply won't offer the benefits coinciding with DirectX 10 specifically.

The improvements inherent to DirectX 10—faster drivers, granular memory access, and unified hardware shaders—should play a prominent role in accelerating game performance. And because the Vista desktop requires 3D acceleration for the first time ever in its lush Aero incarnation, those enhancements also stand to make an appreciable impact on overall system speed, as well. Graphics vendors are understandably excited that, for once, a new operating system is a reason to upgrade.

The GeForce 8800: Making a Transition

Double-Barreled Horsepower
At launch, NVIDIA's GeForce 8800 GTX supports SLI in its usual dual-card mode. However, an extra SLI connector onboard suggests that NVIDIA has plans for even more multi-card madness.

Microsoft claims that consumer Vista is right around the corner, citing a January 2007 timeframe. With the much-anticipated operating system imminent and Redmond promising free Vista upgrades to customers who buy Windows XP boxes this holiday season, you'll most definitely want to lay a Vista-compatible hardware foundation. At the mainstream level, that means sticking to a DirectX 9 board from either AMD or NVIDIA, at least until one of those two announces plans for budget-conscious buyers interested in DirectX 10. Enthusiasts, however, get to play with true DirectX 10 gear today.

NVIDIA's GeForce 8800 GTX and GeForce 8800 GTS represent the first two DirectX 10 boards on the market, each armed with very high-end specifications and a number of brand new features. The pair is designed to accelerate a lot of the functionality bundled into Windows Vista, yes. But they also turn out record-breaking performance results in today's games thanks to an unconventional architecture that is really paying off for NVIDIA.

The pair centers on NVIDIA's "G80" GeForce 8800 graphics processor, manufactured very conservatively on a 90nm process with an incredible 681 million transistors. Just one look at a block diagram of the GPU reveals that it's nothing like any of NVIDIA's past designs. Instead of a traditional graphics pipeline, the GeForce 8800 is loaded with little stream processors—scalar, multi-purpose engines used to operate on vertices or pixel data. They run at 1.35 GHz, asynchronous from the rest of the processor, enabling tremendous floating point horsepower.

Unified Shading
NVIDIA's GeForce 8800 GTX processor consists of 681 million transistors, which comprise the world's first unified shading architecture for the PC. When DirectX 10 emerges, this chip is ready.

In its most complete form, NVIDIA's GeForce 8800 sports eight sets of 16 stream processors, totaling 128. Those eight clusters connect through a crossbar switch to six ROP (raster operation) partitions, in turn attached to a single 64-bit memory interface. Do the math and you get an unconventional 384-bit bus. Though not a very PC-friendly number, there's no tricky math involved, just six independent, 64-bit pathways.

Because the GeForce 8800 is already so big and complex, NVIDIA chose to break its display pipeline out into a separate chip, which sits alongside the big GPU. The separate ASIC features TMDS logic capable of fielding a pair of dual-link DVI outputs and the RAMDACs necessary for analog output. HDCP support is, of course, a standard feature now that Vista is so close.

On Hand and Available

At least initially, NVIDIA is using its GeForce 8800 to power two very high-end graphics cards: the GeForce 8800 GTX and the GeForce 8800 GTS. The flagship GTX features every GeForce 8800 feature enabled, operates at 575 MHz, and sports 768MB of GDDR3 memory clocked at 900MHz (1,800MHz effective). The board is physically massive, measuring about an inch longer than AMD's already-gigantic Radeon X1950 XTX. Obviously, resellers need to keep those dimensions in mind before aspiring to cram the board into a svelte SFF chassis.

The 8800 already represents a massive deviation from any architecture NVIDIA has ever designed, so the company chose to stick with TSMC's proven 90nm lithography process for manufacturing. Consequently, the chip is large and consumes considerable power—so much so that the GeForce 8800 GTX requires two, six-pin auxiliary power connectors. Customers who already own a beefy SLI-compatible power supply and want the GTX should have no problem making the connection. But it's important to remember that the card also supports SLI, as well, necessitating a fairly hardcore power supply. The only two PSU ratings currently certified for GeForce 8800 GTX SLI operation are 850W and 1,000W models. Conversations with power supply vendors during the past month indicate such high-output supplies will become more common in the months to come.

On the subject of SLI, NVIDIA's GeForce 8800 SLI features two edge-connectors instead of one. NVIDIA hasn't yet said what the second connector is to be used for; however, all signs point to configurations with more than two GeForce-based boards. The new 680i SLI-based ASUS Striker Extreme motherboard (see this month's Easy Upsell column) not so coincidentally boasts three x16-length PCI Express slots.

The other GeForce 8800 card bears a GTS suffix and is physically similar. Built on a somewhat smaller PCB and scaled back a bit, the GTS should be paired to a 400W power supply feeding its single 6-pin auxiliary connector. Onboard, the GeForce 8800 GPU is the same. Two of its stream processor clusters are disabled, though, yielding a total of 96. One ROP partition is also turned off, resulting in a 320-bit memory path populated by 640MB of memory at 800 MHz.

Both the GeForce 8800 GTX and GTS are shipping with approximate prices of $599 and $449, respectively. Pricey though the cards may be, your customers can expect performance from one card that, in many cases, exceeds what they'd get from a pair of prior-generation flagships priced similarly.

An Architecture to Last

Performance isn't the new GeForce 8800's only marketable value. There's also DirectX 10 compliance to consider along with enhancements to image quality and video output through NVIDIA's standalone display chip. Both the GeForce 8800 GTX and GTS feature two dual-link DVI outputs, video out, and HDCP support for high-def video playback.

Even if mainstream buyers look right past the GeForce 8800-series, make sure they're at least aware that NVIDIA and AMD both have less expensive DirectX 10 architectures on the way. And since we only have scant weeks until Windows Vista drops, you'll want to set them up with the best possible user experience.