The Green Machine NVIDIA’s flagship, the GeForce 7800 GTX, delivers unprecedented performance in a single-slot solution. Thanks to considerable architectural fine-tuning, the card also runs quieter its predecessors.

In times past, our desktop graphics cover story would have focused on the latest GPUs from ATI and NVIDIA, and that would have been that. This time around, we found that the subject of desktop graphics has greatly broadened. 3D performance is still important, but add-in boards (AIBs) are no longer the only shining star in the sky, and video is starting to eclipse graphics as the killer app for most GPUs. Windows Vista support is the mantra on everybody's lips, and there are now a couple of serious, channel-friendly upstarts on the scene chomping for a bite of the desktop graphics pie.

In short, the rules of selling desktop graphics are changing. It's no longer just about speeds and feeds, which is great news for system builders, because the more complex the GPU's job becomes, the more opportunity you have to educate customers and help them build custom solutions around these new function sets.

NVIDIA's New Era
The GeForce 7

We're not going to descend into statements about which GPU is the best. As of this writing, ATI's X1800 has yet to hit streets or our test bench, making any meaningful comparison between that and NVIDIA's current flagship, the GeForce 7800 GTX (code name: G70), pointless. For now, let's just say that NVIDIA planted its latest graphics generation on store shelves first, thanks in large part to its decision to manufacture on the 0.11-micron node. ATI opted to make the inevitable leap to 0.09-micron (90 nanometers) and lost valuable months mastering the smaller process's learning curve. Apparently, NVIDIA gave the 0.11-micron node one extra fling to buy itself time to master the 90 nm complexities behind the scenes. We hear from NVIDIA that the company has 90 nm discrete parts ready to ship soon, and the GeForce 6100 family (see below) is already 90-nanometer.

First in the GeForce 7 line, the 7800 GTX, by far the most complex GPU of its time with 302 million transistors, launched on June 22nd last summer with immediate channel availability. The leading architectural change from the GeForce 6 generation is a jump in the number of pixel pipelines from 16 to 24, and each pipeline has been optimized to better accommodate longer shader programs. Vertex pipelines stepped up from six to eight. However, the number of raster operation pipelines (ROPs, which are responsible for reading and writing depth and stencil operations) remained unchanged at 16. This would seem to indicate a shifting emphasis toward doing more computation on each pixel rather than boosting the total number of pixels output. In other words, quality is starting to matter more than quantity.

Of course, the GeForce 7 series pushes up the bar on performance and benchmark results, but perhaps even more important are the subtle advances NVIDIA made in its CineFX 4.0 and Intellisample 4.0 engines. Higher levels of antialiasing and filtering are now possible, and the high dynamic range (HDR) effects of the 7 series are stunning. In one example we saw, bright light shines through a window of a dim cathedral. Without HDR, increasing the intensity of the light results in massive blooming that obliterates detail around the window and, with continued increasing, much of the total scene. With HDR, blooming is greatly minimized and instead the scene is evenly illuminated, much as you'd expect to see it with your eyes in real life.

The 7 series continues the use of Shader Model 3.0 that debuted to considerable criticism in the GeForce 6 series. The comparisons NVIDIA provided illustrating the benefits of 3.0 all weighed against examples from Shader Model 1.1, and critics seized the moment to point out that Shader Model 2.0 could achieve just about anything you could do in 3.0, so NVIDIA sure must have made the jump as a marketing move. The mistake NVIDIA made was not illustrating the key advantage of 3.0 over 2.0: time savings. (Admittedly, this is hard to convey concisely to someone outside of the development community.) As Steve Sims, senior product manager, desktop GPUs for NVIDIA puts it, SM3.0 "allows you to program in a way that is far more efficient, and more efficient code runs faster."

Do Seats Do That? While the “nearest competitor” (top)is likely not using Avivo technology, the comparison does show how PureVideo processing can improve 3:2 pulldown errors. If you watch the video of this benchmark demo, the non-PureVideo seats actually appear to ripple.	Read Between the Lines While not as offensive as, say, flag burning, poor spatial-temporal de-interlacing takes a toll on video imagery with high-contrast edges in motion. NVIDIA here shows how its PureVideo technology can help restore glory to the ol’ Stars and Stripes, particularly the Stripes.

Other benefits of the 7800 include HDTV output and H.264 hardware-based decoding assistance. According to NVIDIA, one 7800 GTX is faster than two 6800 Ultras running in SLI mode. The new card is also 10dB quieter than a 6800 Ultra and only requires a 350W power supply, which testifies to the advances NVIDIA has made in performance-per-watt with its new architecture. And, as we've detailed in RAM previously, NVIDIA is no stranger to aggressive channel marketing, programs, and product availability.

"NVIDIA is totally kicking ass in the reseller channel," says graphics industry analyst Jon Peddie, president of Jon Peddie Research. "When R480, the X800 and X850, came out, ATI couldn't build enough of them for various reasons, some of which were not their fault. ATI was able to satisfy its OEM orders but at the expense of retail channel, and NVIDIA moved in with the 6800 Ultra. History repeated itself when the R520 didn't come out. NVIDIA stepped in and said, ‘Here's the 7800, help yourself.' To put it in perspective, that market is only $3 million to $5 million a year. The thing is all the press coverage those products get."

PureVideo

The GeForce 7 series, like the GeForce 6 before it, also sports a collection of features collectively known as PureVideo that perform advanced processing for improving digital video quality. PureVideo combines hardware and software to enable improved capture and display of HD and SD video on PCs with greatly reduced CPU overhead. This aids in tasks such as DVD playback and video rendering. The feature set is similar to ATI's Avivo, although since Avivo hinges on support found in the X1K generation (which we've been unable to test as of this writing), we can't judge the results of PureVideo against its peer. NVIDIA states that PureVideo will soon become pervasive across its product line.

To support PureVideo, NVIDIA uses three video processor engines in each of the GeForce 6 and 7 GPUs. The architecture also allows for the use of shaders to accelerate some of the post-processing effects. All modern NVIDIA graphics cards have the capability to connect to both digital and analog televisions, although some board vendors opt not to implement all of the connectivity. On the software side, there's a decoder that's needed to decode all of the MPEG-2 content. Actually, it's the same decoder at the heart of NVIDIA's famed NVDVD application, although users can also select a third-party decoder, such as the one built into InterVideo's WinDVD.

"Some of the benefits you get from PureVideo include hardware decode acceleration, number one," explains NVIDIA's Patrick Beaulieu, PureVideo product manager. "We currently offer MPEG-2 and WMV decode acceleration. It reduces the CPU from, let's say, 100%, which means you may not be able to play HD, down to approximately 25 percent. That's very important. And the nice thing about our technology is it's programmable, so as new standards become more popular, chances are that we'll be able to program the microcode inside of those GPUs to allow the new standards to decode and accelerate them. We haven't made any announcements on other formats yet, but it's fair to expect we'll have more coming."

That's corporate-speak for H.264 decode support. NVIDIA is still staying vague on this point, and our guess is that it boils down to two things: 1) How much burden can PureVideo take off the CPU for H.264 decoding, and 2) will the GeForce 6 series be able to handle the task or only the GeForce 7? Until Blu-ray and HD-DVD arrive, the company has no reason to commit to a firm answer.

PureVideo also delivers outstanding results in high-definition de-interlacing—a critical process wherein interlaced TV video is adapted to progressive display on a PC—as well as factors like bad edit correction and inverse telecine correction. NVIDIA's PureVideo scaling engine helps enlarge and smooth out Web-based video streams so they don't look like blocky garbage, and there's even a free plug-in for Windows MCE 2005 you can provide to customers that helps optimize the system for HDTV displays and PureVideo playback assistance.

NVIDIA's Born-Again
Motherboard Solution

When NVIDIA abandoned integrated graphics in its chipsets starting with the nForce 3 generation, speculation about the reasons ran all the way from cost savings to enable a single-chip solution to the company simply spending too much brain power on the Xbox to keep up with IGP advancement. Whatever the reason, ATI and Intel have had the mainstream IGP market all to themselves—until now.

Termed "GPU motherboards," NVIDIA has updated and retained its nForce 4 MCP chip and now paired it with a GeForce 6 series GPU. The traditional northbridge/southbridge nomenclature still applies here since NVIDIA has modified the 6100 GPUs to handle interfacing with the CPU and PCI Express lanes. The MCP handles most other tasks, including I/O, networking, and audio.

There are two MCP parts: the nForce 410 and the 430. The 410 is much like the vanilla nForce 4 chip, offering 10/100 LAN, two SATA/300 ports, RAID 0 and RAID 1, and MediaShield (NVIDIA's RAID software). The 430 adds Gigabit Ethernet, another two SATA/300 ports, RAID 0+1 and RAID 5, and ActiveArmor, NVIDIA's integrated firewall technology.

Similarly, there are two GPUs: the GeForce 6100 and the 6150. The 6150 differs from its partner by supporting HD video, a TV encoder, DVI, dual-head output, a 50 MHz clock increase (475 MHz vs. 425 MHz), and two x1 PCIe slots instead of one. Obviously, for a multimedia system, you want the 6150.

"In terms of technology, the 6150 has two pixel pipes and one vertex pipe, which is half of the desktop GeForce 6200," says NVIDIA's Del Rizzo. "The flip side is that the 6150 has the PureVideo technology from the GeForce 7 series, so it actually has more advanced algorithms than the 6200. You're going to get better video playback on the 6100 series than the 6200 in terms of CPU utilization and video quality. That's why we're less focused on the gaming aspect than the video, which is what we think is driving the industry today."

Currently, NVIDIA is only offering GPU motherboard designs for the AMD platform, pitting the 6100 family against ATI's RS480. Unlike the RS480 (or Intel-based RS400 group), which does not support Avivo, these GeForce motherboard chips support HD video via the PureVideo processing engine. The ability to mix and match the GPU and MCP chips allows motherboard makers greater flexibility in targeting given market segments as well as system builders more ability to customize boxes to meet end-user interests. And like ATI's Xpress 200, NVIDIA's motherboard architecture now can keep the integrated GPU enabled alongside a PCI Express card, facilitating quad-head, or theoretically even six-head, display output.

Return of the NVIDIA IGP NVIDIA has come back from its IGP sabbatical in a big way. The 6100/6150 northbridge chips may not reach the graphics performance of a 6200 card, but they do support PureVideo, making this one of the most compelling media center platforms available.

"Let's be honest," says Del Rizzo. "Mainstream users are a different breed. The games they're playing are not the same games as in the high-end. Some of the best-sellers in the mainstream segment are things like Sims 2 and Lego Star Wars, and the GeForce 6 built into these motherboards is more than sufficient to play those types of games. The system builder community should look at this as an opportunity to do promotions based around these top-selling mainstream titles. "Narnia" and "Chicken Little" are coming out this fall, and we expect those to be top-sellers, as well. Mainstream users aren't just surfing the Web. They're doing digital photos and putting together home movies. They're using Adobe Acrobat and things like that. These are mainstream apps that require the use of a GPU. The mainstream user today is a digital media enthusiast, and the GPU needs of this segment are only going to increase next year with Vista. And yes, the 6100 series of GPUs are Vista-compliant."

NVIDIA's GPU motherboard partner list reads like a who's who of the AMD world, and so far all of the designs emerging from Taiwan are microATX for more options in home entertainment or corporate presentation settings. Gigabyte, for example, has a model featuring an external breakout box with S-Video, composite, and component jacks. Ever seen component support out of the box on a motherboard? Neither have we.

...more