SLI Steps Up

As if the emergence of 6600 SLI cards wasn't enough of a clue, the standard dual x8 SLI architecture has now officially descended from the enthusiast sector to the mainstream. SLI boards today sell for as low as $85. Look around the market and you'll find that sub-$100 for a dual-slot graphics board is remarkably cheap and represents a great performance and upgrading story for budget-minded buyers.

SLI: Good News x 2 Want to sell even low-end to mainstream buyers on performance and upgradeability? SLI motherboards are now well under $100. Want to give your enthusiasts something new to drool over? Try NVIDIA’s new SLI X16 platform.

"Mainstream doesn't get a lot of ink because it's not as sexy as the high-end stuff, but from a market and volume position, that's definitely a place we need to play in," says George Myers, product manager for nForce 4. "So over the last six to nine months, we've been making sure that all of our SLI technology was moving down to the mainstream. In August, we introduced a new concept called SLI X16. It essentially enabled motherboard makers and system builders to build systems with two dedicated x16 slots. So we were no longer taking a single x16 and splitting it into dual x8s. We're actually having dual x16s on-board for a total of 40 or 38 PCI Express lanes in the system, depending on the Intel or AMD market. These are features normally found on the nForce Professional side with workstations and servers."

Now, it's no secret that video bandwidth is only barely scratching at the ceiling of AGP's capacity, never mind the fat pipeline of an x16 PCIe connection. This would explain why Myers notes only seeing "on the AMD side anywhere from 3% to 10% improvement depending on the configuration and the application." However, Del Rizzo comments that he has seen a 67% improvement in one certain Far Cry configuration.

With conventional SLI pushed down to the masses, where graphics performance is not always a top priority, NVIDIA is now stressing to system builders that the second x16 bus offers considerable future-proofing flexibility as x4 and x8 adapters for storage and other applications become more prominent.

ATI Arrives In Style
The X1000 Series

As mentioned above, ATI faced many challenges in making the process jump to 90 nm. Signal distribution across the more densely packed silicon did not behave as simulations said it would. The solution wasn't particularly complex, but it took ATI a long time to find it. This is one of the main reasons the Radeon X1000 line now faces a winter 2005 rollout instead of a summer debut.

Razor-Sharp Performance That's pretty sexy, and the Radeon X1800 card isn't bad, either. The new two-slot flagship spearheads ATI's new Avivo platform, introduces a new memory architecture, and even sports somel features new to mainstream PCs, such as Dual Link DVI output.

Hopefully, the flagship Radeon X1800, followed by the X1600 and X1300, will be available through distribution by the time you read this. A very limited number of units were handed to online reviewers in October, and a little Googling will show you their initial results, even though driver support isn't set to be complete until November/December.

The X1000 (X1K) family boasts several new features, including enhanced antialiasing, Shader Model 3.0 support, a new "ring bus" memory architecture compliant with GDDR4, and support for ATI's Avivo video and display platform (detailed in this month's ATI focus insert). For those who plan on handling ultra-high-resolution desktops, the X1K family supports dual Dual Link DVI interfaces, although board manufacturers may choose not to implement the feature. The flagship X1800 XT ($549) uses 16 pixel pipelines, eight vertex pipelines, 512MB of GDDR3 memory, and a 625 MHz core clock. So far, the family bottoms out at the X1300 HyperMemory ($79), which delivers four pixel pipelines, two vertex pipelines, 32MB of memory (128MB shared with system memory), and a 450 MHz core clock.

ATI has taken great pains to redesign its pixel pipelines, in particular using an "ultra-threading dispatch processor" within the pixel shader engine to minimize execution latency. The X1800 can accommodate 16 pixels per thread, and the platform allows for hundreds of simultaneous threads across multiple cores. The ring bus sweeps away the traditional architecture of routing 256 wires from the GPU to memory, which is prone to scaling problems and cache misses, and instead implements a sort of spoked wheel topology around the GPU. The outer rim is a pair of 256-bit rings—which is why ATI calls this a 512-bit architecture—one flowing opposite the other. There are four "ring stops" on the rim, each of which links to a pair of memory modules. Through effective management of data in the GPU's memory controller, the ring bus allows for more efficient fetching of data not being held in cache. Early screenshot samples also show ATI's adaptive antialiasing techniques found in the X1K family continuing the company's superiority in filtering over the competition.

The effects ATI packs into the X1K family are stunning, particularly the high dynamic range rendering with AA, dynamic soft shadows, and physical simulation of rain and water. To prove this last point, ATI created a "toy store" demonstration video showing a night-time scene of a taxi driving past a toy shop window during a downpour. When we first glanced at this, we thought it was live-action video. The realization that this was a CG scene took several seconds to sink in and marked the first time we'd been fooled into thinking PC graphics were real video, short-lived though the illusion was.

More on Avivo

The X1800's use of Avivo shouldn't be understated as the GPU's processing power and its applications to video promise substantial benefits in the months to come.

"When I sell Avivo, in some sense I'm selling futures," says ATI director of North American channel sales Larry McIntosh, "because you're mainly going to see a lot of benefit around H.264 playback—the Blu-ray and the HD-DVD formats. You clearly need a graphics processor to do a lot of this playback work. H.264 is so intensive that there's no known Intel or AMD CPU that can do it. Our X1800 is the first to play back H.264 content very well. You can also see more vibrant, rich colors."

ATI's move to 10-bit processing is central to the X1K's color capabilities. One of the reasons graphics professionals still often prefer to work on CRTs is that the monitors have no trouble displaying 10-bit, 1 billion-color images. Most LCDs, in contrast, only offer 8-bit, 16.7-million color output. This is why photographers will sometimes notice color banding when displaying RAW images on an LCD.

"Avivo in the display engine has a dithering capability," says ATI's Vijay Sharma, manager of desktop product marketing. "We dither the output for a 6- or 8-bit display. So you get a pretty good approximation of one billion colors. With a 10-bit image on a standard notebook or desktop panel, you'll see banding, but if you dither it the banding goes away. And it does that in hardware, so there's no penalty in performance."

Avivo also offers 10-bit gamma correction on a per-color basis. With any PC, you generally have a gamma correction curve in the Properties area, but moving the curve manipulates all three colors simultaneously. With Avivo, you can alter each color separately. This offers game developers, for example, the ability to write their own gamma profiles for their games, and that profile can change throughout the game depending on whether it's day, night, fluorescent-lit, or what have you.

As described in our insert, Avivo encompasses five stages in the video pipeline: capture, encode, decode, post process, and display. The first two of these stages are currently handled by ATI's Theater 550 Pro chip, found on the company's TV Wonder Elite card and other products. The last three stages are tackled by the X1K family.

"One piece of technology we took from our consumer electronics group and its Xilleon chip is the TV encoder," says Sharma, "which we put into our 1000 series. So you've got actual consumer-grade TV output from the 1000 series. This ties into the whole HTPC concept. You can now put a TV in the living room and be pleased with the TV output."

Xpress 200 Finds Its Niche

ATI beat NVIDIA to market by several months with its current IGP family, dubbed Xpress 200, which debuted on the Intel platform (RS400) before quickly landing on AMD (RS480). This turned out to be a two-edged sword. Predictably, the RS400 handily outperformed the Intel 915G on graphics, and the early board samples we saw indicated that ATI was stuffing its BIOS with every imaginable tweak, positioning its platform to be an enthusiast's dream. While southbridge performance, seemingly ATI's Achilles heel with chipsets, was nothing exceptional, at least it was sufficient and stable. Once the 945G arrived and the Xpress 200 still showed graphics superiority, it seemed that ATI had a runaway hit on its hands.

Xpress Your Sales TI’s Xpress 200 chipset is surprisingly robust, enabling everything from low-end budget boards to high-end enthusiast screamers. The chipset is so popular that Intel just selected it to replace its “out of stock” 915G.

Indeed, though OEM channels, the Xpress 200 has been a glowing success. As we've detailed in RAM previously, the graphics core is based on ATI's PCIe X300 part, which brings with it the VideoShader, FullStream, SurroundView, integrated TV encoder, and other enhancements from the discrete side as well as SATA, RAID 0 and 1, Gigabit Ethernet, and similar contemporary desktop specs. What Xpress 200 was too early to get was Avivo and Shader Model 3.0. In some circles, this may be an issue, but the platform still remains an excellent solution for mainstream and budget consumer systems.

Is It Live Or Avivo? ATI’s Toy Shop demo video illustrates the awesome potential of the X1000 GPU family. The HDR, shadowing, water computation, and texture detail are so convincing that, we thought we were watching film, not CGI.

Having seen the capabilities of ATI's reference boards, we expected to see a barrage of enthusiast motherboards run to market based on Xpress 200. (Sapphire's recently released Pure Innovation is the first and only one we've seen so far.) Instead, nearly all of what we've encountered are budget boards leveraging the IGP but omitting a large number of Xpress 200's other benefits in order to hit low price points.

"The fact that we had such a good IGP product may have hurt us in the sense that a lot of the ODMs who were adopting the ATI chipset put the focus on building value-oriented integrated graphics platforms, and in that we've been very successful in those markets," says ATI's Niles Burbank, group product manager for chipset products. "But that may have taken the focus away from using Radeon Xpress in other applications where it would be equally suitable, such as the enthusiast segment."

With the Xpress 200, ATI has now shipped over 40 million core logic chips and has more than 40 design wins in the channel. ATI currently ranks first in the AMD platform PCIe IGP market and second behind Intel on the Pentium side. And in a bizarre twist that conveniently coincides with an Intel 915G parts shortage, it now seems that Intel will adopt ATI's forthcoming RC410 IGP part on its low-end motherboards.

Even though it will offer graphics on par with the X300 SE, RC410 still trounces the 915G and at a notably lower price point. "RC" chips differ from the "RS" line in that the lower-end part uses single-channel memory architecture to lower system ASPs by only requiring one DIMM. ATI's documentation notes that RC410 supports all Intel desktop CPUs, but we confirmed just before press time that Intel's implementation does not support the dual-core Pentium D (recall that the 915G only runs single-core), not that a lot of dual-core buyers are likely to adopt a single-channel IGP solution. More importantly, ATI states that RC410 meets Windows Vista requirements today, and we're guessing that the 915G does not. Hence it behooves Intel to jettison any non-compliant graphics cores ASAP without giving up its industry-leading grip on low-end boards.

CrossFire Gaining Speed Expect CrossFire cards and motherboards to start hitting their stride for the holidays. ATI’s dual-slot strategy should appeal to those wanting to balance performance with amazing visual quality.

"That's going to be the wake-up call to the channel," says ATI's Sharma. "If people are selling sub-par or less capable integrated graphics, when people upgrade to Vista, their screen is going to look different than the screen they saw at work or at a friend's house, and it's going to be because of graphics hardware. It's best to get ready for that now, or at least from January onwards."

...more