RAM | Reseller Advocate Magazine: Issue 69

By John Martinez

We’re long-time proponents of the platform approach to system building. Intel’s Centrino proved years ago that there’s value in delivering a total solution rather than harping on speeds and feeds or this versus that component.

Clearly, intel also saw the genius in its ways, because everything we’ve seen from the company since Centrino has had some sort of system-wide spin on it, from consumer-friendly initiatives like Viiv to the more subtle server platforms, such as Bensley.

System builders and integrators are going head-to-head against tier-ones that run their own lab tests and validate custom designs. That’s why platforms are so attractive in the channel. When Intel defines Centrino, it establishes minimum requirements, compatible components, and approved configurations. Add reseller-specific initiatives like VBI to the picture and it’s easy to understand why you go with Intel when you build a whitebook. Resellers without the resources to do their own validation can combine the pieces of a platform, giving customers the benefit of the time and money that went into designing the platform.

Intel's Platform Mastery

It’s one thing to design platforms that tier-ones can take and turn into profit centers. It’s quite another to coordinate channel-friendly solutions. The Intel chipset-processor-motherboard combination has always worked fairly well. Even when AMD hurdled the NetBurst microarchitecture’s performance, ease of integration remained one of the reasons to stick with Intel.

Today, you’ll find Intel advocating its platforms in three different arenas: on the road with Centrino, in the corporate desktop with vPro, and inside multimedia powerhouses through Viiv. Barebones servers, like the SSR212MC2 “McKay Creek” platform, don’t get as much mainstream press, but they still give resellers the same type of validated platform foundation. The 2U, storage-dense server can operate as a NAS, SAN, or DAS box, depending on the client’s needs. This is flexible platform design at its best.

Intel uses a different set of components for each type of platform. For example, the SSR212MC2 combines dual- or quad-core Xeon processors, an Intel server board, a right-sized chassis, and Intel’s newest SAS storage controller. By adding memory, a customized combination of SAS/SATA hard drives, an add-in PCI Express card, and one of the many certified software solutions, you can differentiate the box in a myriad of ways.

Making Money With Platforms
The latest iteration of Centrino, called the Santa Rosa platform, features a processor, chipset, and wireless networking solution. VARs able to combine that platform with Intel's Common Building Blocks can offer some pretty amazing mobile hardware configurations.

Resellers building vPro or Centrino machines have to abide by certain platform requirements as well. In the cases of both desktop vPro and Centrino Pro, those requirements include the Intel processor, a specific “digital office” chipset with AMT support, the right Gigabit Ethernet controller, and vPro-optimized software.

Intel Charts a Familiar Path

For all of its emphasis on platforms, Intel’s best attempts at designing a desirable graphics core have previously attracted very little attention. They were most often deemed merely adequate, packed onto value-oriented desktop boards and mobile chipsets. But Intel’s newest X3000-series GPU is actually turning heads. For the first time, an integrated GPU is leveraging some of the same features found on pricey discrete cards, like unified shaders. Built-in graphics are no longer a liability for resellers; they’re an upsellable feature able to accelerate the very best desktop settings Vista can muster.

Intel’s next graphics project is still a long ways out, but it goes much further than simple integrated graphics. Apparently, there will be a number of products that center on one architecture, currently code-named Larrabee, which will use lots of parallelism, just like AMD’s and NVIDIA’s current flagships. Although Larrabee isn’t going to be part of your platforms until sometime in 2008, it’s still very much relevant given AMD’s emphasis on co-processors and the recent ATI acquisition. In the future, we might see a battle between AMD’s best video card and Intel’s newest discrete board.

AMD: Ready To Put Up A Fight

I still remember my reaction when I read that AMD was acquiring ATI. It was a bit surreal, and at first I thought it was a joke. What would a very partner-oriented processor manufacturer want with a graphics card company? AMD was apparently thinking on a different wavelength since it was willing to spend $5.4 billion dollars acquiring the company, $2.5 billion of which was borrowed.

Publicly, we’ve always heard AMD insist that it was focused solely on processors and that its partners were able to deliver the best platforms. AMD only developed its own chipsets when it was championing a new technology. The first slot-based Athlons used AMD chipsets, as did the first motherboards with DDR memory. Nowadays, AMD doesn’t have to convince third-party chipset vendors that it has hardware your customers want. The chipset folks are ready to roll right when AMD drops its CPUs.

940-Pin...Co-Processor?
Connecting co-processors to the HyperTransport bus happens two different ways, according to AMD's Torrenza initiative. Either you drop them into HTX slots or an available Opteron processor socket.

With the ATI acquisition, however, it’d be hard to ignore the shift toward an Intel-ish platform approach. For every Athlon 64 or Operton processor the company sells, a motherboard partner gets to push an AMD-compatible chipset as well. Adding desktop core logic, graphics, and a mature mobile infrastructure improves AMD’s credibility amongst the customers who’d buy the carefully planned-out Centrino, Viiv, and vPro platforms.

Up until now, AMD has only reached out to the channel a couple of times, and then only feebly. We’ve asked repeatedly about the mobile Turion and how resellers could get their hands on it. We’ve also heard plenty about LIVE! and the Quad FX programs. While executives at AMD have assured us on several occasions that the company had grand plans for the channel, we’ve seen very little VAR involvement in any of its initiatives. At the end of the day, it seemed that exclusively emphasizing processor technology hurt AMD’s ability to evangelize broader efforts. ATI’s entrance onto the scene makes it much easier to imagine notebooks based on AMD logic. There will be a far better chance for AMD to realize success in consumer electronics. And it isn’t too hard to picture a real enthusiast platform based on multi-core CPUs and graphics cards in a CrossFire configuration. So while AMD might have shied away from Intel’s approach in the past, its new position is very much about the strength that comes from a complete portfolio of products that, together, give you compelling solutions to sell.

CPU/GPU Convergence

One thing we’ve been noticing lately is the desire from CPU vendors to do the job of graphics folks by promoting parallelism. From the other side, the graphics folks have pushed to do more general-purpose work using the parallelism and programmability already built into latest GPU architectures. It’s a little tough to tell which way this one is going, but AMD certainly bolsters its position by picking up ATI.

Just take a look at AMD’s recent ATI Radeon HD 2000-series launch. In its press material, AMD talks about the advancement of accelerated processing and the eventual inadequacy of homogeneous multi-core CPUs. Platform and silicon-level acceleration are, it says, the next major inflection points. The physics engine in the ATI HD 2900 is one example of GPU-based acceleration tackling traditional CPU tasks.

The physics engine isn’t a one-off, either. One year ago, AMD announced the Torrenza initiative to get more hardware acceleration integrated onto its platforms. Torrenza specifically involves connecting co-processors to the HyperTransport links already available on Athlon and Opteron CPUs. The initiative is specific to external acceleration from devices plugged into HyperTransport Expansion (HTX) cards and co-processors that are pin-compatible with AMD’s socket interface, but it also includes PCI Express-based graphics adapters like the Radeon HD 2900.

What’s this ecosystem of co-processing going to accomplish for AMD? Beyond the dual- and quad-core designs we’re already seeing, what is going to keep your customer buying new CPUs? Scaling clock frequency is already difficult in light of the sheer complexity today’s chips incorporate. And why would an SMB want to spend more money upgrading to 8- or 16-core chips when today’s software is still being optimized for threaded execution? Adding more cores right now simply wouldn’t make much sense.

Re-Purposed Power
Today's ATI Radeon 2900 HD is a super-fast graphics card. Add professional development tools and an initiative like AMD's Torrenza, though, and you might see it handling higher-end processor loads soon.

But by acquiring ATI (and its Radeon HD 2000 series), AMD picked up the hardware it needed to help develop a platform rich in acceleration. Of course, that’s just half of the story. AMD is also building a software environment consisting of its own Compute Abstraction Layer libraries and compiler extensions that let software developers use their existing programming tools to target stream processors. AMD already has a long list of applications able to benefit from utilization of its graphics chip’s stream processors. Most are higher-end: direct volume rendering, isosurface extraction, and medical visualization, for example. However, mainstream users may find themselves transcoding high-definition video using GPU acceleration in the near future.

So Who's the Odd Man Out?

You could argue that ATI’s role in the graphics market has not been one of dominance for the past year or so. Though NVIDIA has its slip-ups, the company is usually back on top after one generation of enduring second-best. On its own, ATI would have had to continue struggling against NVIDIA for market share in almost every facet of its business. Teaming up with AMD gives it a strategic advantage thanks to the new platform approach. Hopefully we’ll see better chipsets, a manufacturing arrangement that adds value to ATI’s GPUs (like NVIDIA, ATI was fabless), and compelling integrated graphics able to do battle with Intel.

You can be sure that Intel, the innovator in platform technology, isn’t going to be looking for an arrangement similar to that of its main competitor. The company already has a booming processor business and is known for its quality chipsets. The one area where it has historically shown weakness—graphics—is now getting a lot of attention. The likelihood of Intel looking for an acquisition of its own diminishes as its graphics portfolio gets stronger. If Larrabee proves successful, Intel will be a player in the discrete graphics market.

With AMD and ATI in bed, and Intel content on its own, NVIDIA looks like it’s standing alone. Outwardly, NVIDIA is probably thrilled. It remains on top of the desktop graphics market and, in at least the near future, there’s really nothing AMD can do to improve ATI’s GPU lineup. In fact, AMD’s most recent Radeon launch lacked the organization and precision of debuts past. NVIDIA’s nForce brand is also the biggest thing in chipsets for AMD. Without nForce, a lot of customers currently buying Athlon 64 processors might otherwise pick up Core 2 Duos, which AMD doesn’t want.

Think about the future, though. Both AMD and ATI come to the table with solid technology and complementary offerings. Once they fully integrate and start firing on all cylinders, third-party chipset vendors will cease to be as relevant. The platform story AMD is looking to tell—the one Intel has so much success telling today—involves processors, chipsets, graphics, and networking all coming from one vendor. At that point, NVIDIA may find itself looking for a dance partner.

Sticking to the Guns

NVIDIA’s biggest strength remains its graphics processor business. The top GeForce card leads its target market, and the fastest Quadro also dominates the professional pack. But based on the most recent advancements made in Intel’s integrated graphics lineup (and murmurs from the Intel Developer Forum), a discrete card from the CPU giant seems inevitable. With AMD and ATI working to develop stronger platforms on one end and Intel gearing up to get busy in graphics on the other, NVIDIA may have to fight for relevance in the channel.

If anything, NVIDIA’s saving grace will be the professional Quadro FX cards. Although the Quadro line is based on the same technology NVIDIA sells to gamers, nothing in ATI’s repertoire comes close to the performance that NVIDIA offers. As for Intel, its vision in the professional space is still a mystery. At the high-end, cards like the Quadro FX 4600 stand uncontested, and NVIDIA’s Quadro Plex visual computing systems ratchet up performance from there.

Only time will tell how NVIDIA fares against Intel’s best effort in graphics and AMD’s newfound faith in platforms. Intel is still early in development. AMD seems to still be working out the kinks of its latest acquisition. But no matter what happens, the channel stands to win big from an almost universal push to accelerate computing at a platform, rather than processor, level.