There's an emerging theory that the central processing unit is becoming less and less responsible for defining what a PC truly is, and therefore, who it sells to. Two other components are shedding their ancillary roles, and stepping up to define how powerful a system can truly be: the chipset and the graphics processor. AMD last week placed a huge gamble on this theory being correct.
As ODM customers continue to drive the mobile computer market toward a role as the dominant segment, IDC's Shane Rau believes it will take on some characteristics of the embedded device market, much like handsets today. In such a market, differentiation will be achieved through a kind of segmentation that leads to specialization - devices that distinguish themselves from one another through their function, and that are less general-purpose than they are today.
"That's one of the decisions that we just haven't made: What are we going to do with the ATI Radeon graphics brand?"
Rick Bergman, Senior Vice President, ATI
The way Rau sees it, there are now a total of five serious players in the chipset arena: Intel, Via Technologies, SiS (for as long as it hangs on), Nvidia, and now the combined AMD/ATI. Market forces being what they are, the drive towards consolidation could conceivably push this number down to three, with Via actually surviving, perhaps thriving. "I think Via was probably the first company positioned for a platform approach," said Rau, "because they brought in all the silicon, and talked about connected computing years before we saw our first products from the other companies."
Rau's model foresees the possibility of platform-based devices emerging from the mobile PC space, but not necessarily requiring the mobile aspect. ODMs, therefore, could craft such things as point-of-sale systems and specialized thin clients, using low-cost PC components instead of higher-cost ASICs, but still maintaining many of the characteristics of embedded systems. In such an environment, Rau concluded, Via "could be leading the segmentation...It may be Via who saw this earlier than everybody else, and then started running in that direction."
The other players in Rau's model will likely include Nvidia, with or without a partner. "For any company that wants to survive," he said, "they're going to look for mechanisms to do so. That mechanism is often based on, if you can't do it better than somebody else, then you take what resources you have and apply them in different ways."
What would Nvidia's formula be? Rau projects another take on the combined CPU/GPU approach ATI and AMD alluded to last week, with platforms including - but not limited to - game consoles, where the graphics processor assumes the principal role. "The more specialized [devices] are in terms of graphics, the more the system partitioning may be pushing towards the GPU, where the GPU may be the dominant processor inside of the system, the big honcho that does the heavy lifting," he said. "And it might be a more modest central processing unit to do some general-purpose stuff; or you might see the GPU taking on some of those central processing features, but it's still a matter of graphics being the head honcho in the system."
But market forces being what they are, like a universe that expands and then contracts again, Rau thinks companies will be compelled to align their various technology segments, resulting in the end in the usual duopoly. "I think what probably will play out is, what was happening in multiple, parallel markets - processor, chipset, graphics - is now going to happen across two companies. The Intel/AMD dynamic, as far as processors, will continue, but then the kind of duopoly that was going on in graphics between Nvidia and ATI will continue, except it will be AMD/Nvidia."
In-Stat's Jim McGregor disagrees...completely. "Basically, I can see where they're coming from," he told us, "[believing that] chipsets are key because they're going to determine what key features are in the system, from I/O ports to interfaces to graphics...Yes and no."
Or so McGregor said, but really more no than yes. As he explained, what's really driving the evolution of the PC is consumer usage models. The UMPC is one example of a platform designed around a usage model - it isn't particularly a working example, he concedes, but he believes by its next generation, it will be. "Obviously it's a lot cheaper to do one solution and just not use certain things, than to develop fifteen different chipsets for fifteen different market segments. So I wouldn't go so far as to say that you're going to have a different chipset for each market. I don't think ATI or Intel or anybody wants to build different chipsets for each niche market."
Which is the whole idea of a platform, really, as McGregor explained - a way to prevent segmentation from taking hold across the board, and dividing PCs into glorified embedded systems without the embedding. Manufacturers are looking for the most common solution, he said, to fit the widest possible range of usage scenarios. But to accomplish this, don't forget that integration is also taking place on the intra-chip level. As 22 nm lithography becomes something foreseeable perhaps within this decade, CPU manufacturers will be looking toward how much they can integrate onto the chip itself.
"With the transistor budgets increasing so significantly, are we going to get to a point where it just doesn't make sense to have separate graphics solutions?" asked McGregor. "And especially with a focus on power consumption and performance and everything else...it's getting to the point where, in a lot of cases, it may not. And that's the challenge that ATI and Nvidia both face: Where do they go from here? So it doesn't surprise me [that] ATI was looking for alternatives."
What prevents AMD and Intel from just taking the plunge today and integrating graphics, northbridge, and chipset functions onto the central processor right now, believes McGregor, is the need for the functionality those other components provide to be thoughtfully tried and tested in the field. "Nothing gets integrated onto the processor or the chipset in the first generation." For example, there's talk today about whether Wireless USB should at some point be integrated into a single-silicon solution. But the talk stops just after the point where the question is raised...Nobody wants to be the one to have to give it a try first.
"OEMs who make the systems want to be able to differentiate. They don't want to have all the same silicon in their system."
Shane Rau, semiconductor analyst, IDC
"I would say the chipset plays a key function in terms of determining the system configuration," stated In-Stat's McGregor, "but all those new technologies that are really going to increase the market inflection point are going to be discrete solutions for the first and second generation."
One part of McGregor's model does not differ significantly from Rau's: He sees the chipset market consolidating to as few as four major players, with Via finding its own niche and surviving comfortably. But GPU manufacturers - with ATI still in the mix - must continue to develop their own IP, even if it inevitably becomes consolidated into a general-purpose processor three or more generations down the road. For that IP to develop, it needs a staging ground, a proving field, a place for the market to determine which components truly perform and which don't.
For that reason, McGregor says, the premium graphics card segment can and must survive. "You still have to have a good performing solution, because if you don't, what ends up happening is, then [as a consumer, I might find myself] asking, 'Do I want an Intel solution, or do I want an AMD solution?' Well, if the processors are relatively equivalent, and I want it for this particular segment of the market - say mobile entertainment, 17" mobile computers - I want something that's got real high-end graphics...If the processors are equivalent, it comes down to the graphics solution."
All our wild projections inevitably come around full-circle to the following revelation: Without the enthusiast graphics segment of the market, neither the ATI division of AMD nor the Nvidia division of Nvidia will know what deserves to be integrated within a mainstream platform, and what doesn't. Whether or not AMD has actually come to this realization is not yet clear, but even if it hasn't, unless Nvidia drastically changes its course in the coming months, its own future in the performance segment may actually have just been secured.