|
||
|
|
|
              |
|
 |
 |
||||||||||||
|
 |
|
||||||||||
By William Van Winkle |
||||||||||||
Here we are at the third quarter of a year that’s turning out to be economically tougher than just about anyone predicted. It’s as if we all time warped back five years and suddenly everyone is sitting on their wallets like over-possessive hens. That’s OK. Even in the worst of times—and we’re a long way from that—people still need to address their technology needs. The question is: How are you going to do that in a way that is more persuasive than what your competitors are attempting? |
||||||||||||
What you need isn't just bang for the buck. You also want new functionality enabled at mainstream price points, not lofty early adopter levels. This is a game in which following the status quo may not lose, but it sure won't get you ahead. The object is to boost value quickly, easily, and affordably while increasing usage flexibility. The steps you take to spike up value should yield benefits today and keep on adding more benefits into the future as your customers' needs scale. Perhaps more than any other consumer PC technology, graphics grabs most of the headlines and for good reason. The overarching goal of graphics technology is perfect realism in real-time. When you can't tell the difference between an LCD display and a window, we'll be there, but that day is still a long way off. Purveyors of integrated graphics technology like to point out how far IGP output has come, and to some degree they have a point. NVIDIA with its recent GeForce 61XX northbridge chips in particular have raised the bar for integrated video output without going discrete, and this in part explains why IGPs own such a formidable part of the graphics marketplace. (With the weakest performance and functionality of the three major IGP companies, Intel still commands well over half of the total graphics market, if that puts things into perspective.) But everything stays relative, right? Two years ago, the idea of IGPs handling high-definition content would have been absurd. Now high-definition video is seen as the next weak point for IGPs to for integrated chipsets to conquer. More importantly, for all of the benefits of integrated graphics, IGPs show obvious shortcomings. Two years ago, an IGP playing DOOM 3 (at acceptable resolution and fps) would have been insanely cool, and most late model IGPs can accomplish this. However, now Elder Scrolls IV: Oblivion would leave even the most cranked IGP whimpering in fear. But this is misleading. Graphics has historically been a battle over speed. The scene is shifting now, and elements like physics processing and video optimization are rising up alongside frames per second and image filtering as top priorities. These new functionalities don't always require the fastest, most expensive cards, but they usually fall beyond the scope of IGPs. This is exciting, because at last we have grounds to resume selling value-adds in the graphics world. We're no longer bound to the price=performance equation. And with broader functionality comes more opportunity for upselling and configuration differentiation. Because let's face it—no one makes any margin selling IGPs. There are many new developments in the graphics world that resellers should grasp, but the three most exciting are multi-slot parallelism, physics processing, and video optimization.
Selling Hot with Multi-Slot Being first to market can be a risky affair. On one hand, pioneers get the benefit of concept association (think Xerox for photocopiers and Kleenex for tissue) and loads of media attention. Then again, those who lead the pack in a new direction are generally the ones who get eaten first. Mention "multi-slot" or "multi-GPU" technology to anyone (who doesn't work for ATI) and you're likely to hear, "Oh, sure, you mean SLI." In 1998, 3dfx's Voodoo 2 was the first to implement "scan-line interleaving" using two graphics cards joined by a ribbon cable to perform parallel processing on graphics tasks. When NVIDIA scooped up 3dfx's assets at the start of 2001, SLI came along with the bargain, but it would be a little while until the thermal pinch that plagued desktop processors also hit GPUs. By 2003, it was clear that the insane escalation of single-processor performance would soon moderate in the face of thermal limitations, and parallelism would become the upward path of least resistance. Thus in 2004, NVIDIA revived SLI, newly redubbed as Scalable Link Interface. NVIDIA has played SLI to near-perfection, starting the technology with the same gamers and enthusiasts 3dfx left off with and more recently pushing dual-slot graphics into the mainstream with nForce 4 SLI chipset-based motherboards ringing in under $90. While NVIDIA eventually brought SLI to the Intel platform, the company undeniably hitched itself to AMD's rising star, and SLI quickly became a staple on any consumer machine where graphics performance was the top priority. By now, you know the basic pitch: SLI uses a pair of compliant GeForce 6, GeForce 7, or Quadro FX series PCI Express cards on a motherboard with an SLI-compatible chipset and two (or more) x16 PCIe slots. More specifically, conventional SLI cards use eight PCIe lanes, and traditional SLI motherboards pipe eight PCIe lanes to each of the two x16 graphics slots. Given the massive bandwidth advantage of PCI Express over AGP, NVIDIA rightfully felt no need to see more than x8 throughput to each slot. Certainly, NVIDIA could have pursued a design wherein two GPUs mounted on a single card and operating in parallel shared a full x16 slot connection, but this would have gone against the "buy one now, add another later when you need it" value message surrounding SLI. (Never mind that the significant majority of SLI-compatible systems are sold with two cards already installed.) SLI has a few caveat points. SLI cards almost always feature two display outputs, and this gets you dual-head output from one card or quad-monitor support from two cards in two slots. Unfortunately, SLI only works on one monitor, requiring that the other displays be disabled before SLI will work. At heart, SLI works by having each GPU compute alternate frames (AFR, alternate frame rendering) or by having each GPU compute 50% of each frame (SFR, split frame rendering). Because the work load is equally balanced in both cases, NVIDIA generally requires that SLI cards be a matched pair. Early on, this meant that you needed to have two of the exact same card. More recently, the rules have relaxed and now you only need two cards running the same GPU. For a brief time, there were hacks that let SLI cards run on non-NVIDIA-based chipsets, but NVIDIA modified its drivers and kept all of the SLI glory for itself. If you want SLI, you'll only get it through an nForce 4 or 500 series SLI chipset. Of course, if a little SLI is good, more is better, so NVIDIA announced Quad SLI. Forget four single-GPU cards in four slots, and forget the handful of cards that appeared with two GPUs on one PCB. Quad SLI is ideally suited for motherboards based on the nForce 4 SLI X16 or the nForce 590 SLI, both of which pump a full 16 lanes to each of the two x16 PCIe slots. (Quad SLI also runs on standard SLI boards featuring two x16 slots utilitizing eight PCIe lanes each, although this can yield bottlenecking in some situations.) The other half of the equation is "GX2" cards, presently only available within the 7900 GPU family. Like the Obsidian X-24 card of old, a GX2 is essentially two SLI boards bolted together and linked to share the x16 PCIe connection. Plant two GX2 cards on one motherboard and you've got a full 32 PCIe lanes running at full bore.
Since we're talking about adding value here, be aware that NVIDIA baked a little feature into the 590 chipset called LinkBoost. If you're running a 590-based board with at least one 7900 GTX GPU, then the system's high-bandwidth connections automatically overclock by 25 percent. Thus the HyperTransport link between the SPP and MCP (each of which supplies one of the x16 PCIe feeds, if you're curious) bumps from 1 GHz to 1.25 GHz. Spin this however you like. Sell the high-end parts and offer "free," factory-sanctioned overclocking. Or save your customer some cash with a 7800-generation product and do the overclocking manually. It's good value either way. Now, before you run off and start shouting the Quad SLI message down the street, beware of some first-generation issues. Most reviewers are reporting few if any performance benefits below 2048 x 1536, and even at that level, Quad SLI earns its keep better with filtering cranked up. Moreover, reviewers in the May time frame reported significant instability with the Quad SLI drivers. Just as the initial SLI required a huge rework of the GeForce drivers, no doubt Quad has presented NVIDIA with a whole new set of challenges. That's fine. NVIDIA is among the best at sorting these things out. Just be cautious at this early stage and suggest to buyers—who will easily drop over $1,000 for Quad SLI—that perfect stability may take a few updates. On the other side of the fence, we have ATI, which brought its own belated CrossFire multi-slot solution to market during September of 2005 in the midst of probably its most problem-prone period since the pre-Radeon 9700 days. The company was running late across the board (thanks to massive work on the Xbox 360 project), NVIDIA was pulling ahead in the all-important high-end desktop segment, and when CrossFire finally arrived and made the review rounds, general feedback seemed to revolve around the theme: "That's all you've got?" After a few updates, most agreed that CrossFire and SLI were more or less at parity. ATI's message of greater DX9 universal compatibility, enabled through a third "SuperTiling" rendering approach, fell mostly on deaf ears since all of the hot ticket games coming onto the market included SLI support. The requirement of an external dongle instead of NVIDIA's internal PCB-to-PCB connector was viewed as an aesthetic gaffe, and resolution and refresh rate limits in the first X800 and X850 CrossFire cards made few friends. More annoying was the requirement of a "master" version of CrossFire Edition cards to run alongside the "slave." Fortunately, persistence has paid off for ATI and its followers. As NVIDIA also found with its lower-end SLI SKUs, the bonding of the two GPUs, in this case the X1300 and X1600 parts, could be done across the bus rather than with some sort of bridge. Those using ATI's Xpress 3200 chipset (good: delivers twin full x16 PCIe channels; bad: still only available for AMD processors) don't need to worry about master cards save only in the X1900 range. And with the X1900 family, ATI has taken its turn again at slipping past NVIDIA in the benchmark races. NVIDIA might counter that a 7950 GX2 outstrips an X1900 XTX, but that's two NVIDIA GPUs versus ATI's one. We can't wait until "performance per watt" or some similar new measurement method takes over the GPU space. With the ascendance of the X1900 series comes the corresponding rise in CrossFire's fortunes. ATI scored a huge credibility gain when Intel announced that it's dual x16 slot 975X motherboards now support CrossFire. In fact, 955X-based boards with twin x16 slots are compatible with CrossFire for X800 and X850 cards—a little-known fact that could prove handy for upgrading last year's power user configurations. But the big ATI performance story now is CrossFire on the X1900 family. "Everybody who's tested our 1900 on CrossFire, whether it's on AMD or Intel, says it's tremendous," says Larry McIntosh, director of North and South American sales for ATI. "The challenge is that they've already got a system cooked with SLI and AMD, and they're probably not going to requal a similar solution with different video. What they want to do is transition with a new launch point with a new platform, which Conroe is. The big guys can't do these one-week quals. It's just too expensive on the back side. But by the end of July, you're going to see this big shift to ATI-based systems, first on Intel, and then we'll see a similar shift on AM2."
Like it or not, big OEM designs are crucial to channel success. NVIDIA premiered Quad SLI with Dell back in January, and here were are six months later just starting to see product for system builders. Without the Dell (and subsequently Alienware) buzz, manufacturers wouldn't have seen enough market awareness and demand to justify channel SKUs. Part of ATI's struggle over the past year was that there were relatively few OEM wins for CrossFire, and now that looks to be changing. The difference between Quad SLI and next-gen CrossFire is that ATI's platform arrived in the channel before the big OEM push. Today's Intel X975XBX "BadAxe" motherboard hosts three x16 PCI slots, is fully compliant with X1900-class CrossFire graphics, and will support Core 2 Duo ("Conroe") processors provided the board is revision 304 or later. Thus as a new system or upgrade item, the BadAxe/CrossFire sale makes great sense as a future-proofed performance play. Still, ATI doesn't get to claim all of the next-gen multi-slot glory. NVIDIA promises 590 SLI and probably 570 SLI boards (both chipsets are Conroe-ready) with four PCIe graphics slots as well as boards with three slots. There are two usual ways to approach multi-GPU graphics: speed and quality. As mentioned already, the speed issue largely boils down to the chips and cards populating the slots, not so much the SLI or CrossFire technologies themselves. Even reps at both companies agree that the two approaches now tend to test out very similarly.
And in good marketing tradition, the rivalry between ATI and NVIDIA over image quality superiority is open to interpretation. One of the core tenets in ATI's recent CrossFire gospel is that it offers simultaneous HDR and AA filtering whereas "the competition" does not. On a literal level, this is true. However, NVIDIA has devised a work-around of sorts for titles such as Tomb Raider: Legend and Day of Defeat: Source. Additionally, quad SLI sports a new 32X antialiasing mode and 16X anisotropic filtering, although real-world perceptions of improvements enabled by these updates have been mixed, and few users have monitors with high enough resolution to realize Quad SLI's chief benefits. ATI hasn't made a stink about going beyond two GPUs for graphics rendering, and no one has seemed to mind. No, it seems the real reason to get excited about Quad SLI and moving beyond two graphics slots lies in a different direction altogether: physics. ...more |
||||||||||||
 Back to top
|
||||||||||||
Copyright © 2007 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form. |
||||||||||||
