Round Rock (TX) - AMD's first genuinely successful product in the PC CPU market was a product called K5. Far from its first processor, K5 was positioned as a price/performance competitor to Intel's Pentium. Later, AMD established a position for itself on the high-end as well, and even on the premium side of the price scale, as a performance leader. This was a strategy that would confound Intel's best efforts for nearly two years. But those periods of history are marked by the milestones of Intel's competitive responses, all of which drove AMD to new directions.

Core 2 Duo in detail: Past, present and future (22 images) ...

See the AMD AM2 processor slide show ...

"Historically, if you go back to periods when AMD has gotten above 20% [market share], Intel has responded viciously," related Shane Rau, research manager for the semiconductor group at IDC. "In 1998, it was a product response; that's when [Intel] introduced Celeron. That took back share. In 2001, it was a price war, through which Intel took back share and indeed drove AMD into the red. This time, it's a response based on both price and product. That level of response testifies to the strength of positioning that AMD has today, relative to those prior two periods."

Were it not for AMD, the Core 2 Duo series of processors that emerged from Intel yesterday would not be as strong as they are. But if history is any guide - and as history repeatedly teaches us, it is - then this is the part of the script where AMD digs in, reaches down deep within itself, and pulls out an overwhelming new advance in CPU architecture that everybody sees coming except Intel. AMD can no longer position itself as the "value alternative;" a company can't lead by being alternative.

"The evolution of AMD is such that they became a serious player and innovator based on their ability to produce performance products," Rau told TG Daily, "so that they didn't have to compete solely on price. Now, while still a substantial portion of their desktop chipsets are in the value space - Sempron is upwards of 50-55% of their shipments on a quarterly basis - it's the high-end products, the FX, the Athlon 64, the high end and its ability to compete toe-to-toe with the Intel counterparts on performance, that allow them to be where they are today, to gain market share, to raise their ASP [average selling price], to increase their revenues."

Like Intel, AMD needs a premium processor not only to give itself a high-margin product line, and not only to prove its performance ability, but also to set the trend for the mainstream products which that premium line will inevitably metamorphose into, over the next two or three product cycles - which today is typically about 18 months' time. Today, right this moment, AMD has a harder time justifying its $827 FX-62 as a performance leader, especially when that performance is bested by an estimated all-around average of 25.8%, based on Tom's Hardware Guide tests, by Intel's Core 2 Extreme.

"The critical thing for AMD is, what do they do next?" asks Jim McGregor, principal analyst for In-Stat, and editor of the In-Stat Microprocessor Report. "They've got a really killer architecture. Their recent enhancements [include] adding DDR2, and manufacturing a revision that's going to allow them to do some mid-term enhancements with some frequency increases. So really, what's going to be the huge benefit in that? It's going to be minimal, it's not as drastic as what Intel has gone through."

A more granular scale of advancement

As McGregor explained to us, AMD's advances don't come in huge, thunderous waves like Core 2 Duo. Instead, they come in granular advancements that are smaller, but much more frequent. So "next" for AMD, with regards to the next few months, may be something fairly small and simple. "You have to remember that AMD's architecture still has room to increase frequency," he said. "Where Intel kind of ran up against a brick wall with their old NetBurst architecture, AMD still had room to go."

AMD has essentially been enhancing and revising the same basic AMD64 architecture it introduced in 2003, McGregor reminded us. That's not to say AMD64 is "aging" by any means; it's to say AMD was smart in planning so far in advance, essentially forcing 64-bit and dual-core technologies into the public vocabulary, and forcing Intel to follow. But can AMD go on just tweaking AMD64, without an entirely new architecture that blows Core 2 Duo out of the water?

Jim McGregor believes, yes. The key, he said, is in how it responds to this notion that just because it has yet to complete its transition to 65 nm lithography, while Intel is humming right along at 65 nm today, it has fallen a year or more behind Intel. AMD is nowhere near that far behind, McGregor explained, if you look at the way the two companies engineer their manufacturing processes.

"Intel has kind of a brute-force method," said McGregor, "where they come up with a process, and all at once, they try to transition every product to it, and make the product run on the process. AMD and the rest of the industry (that doesn't have the capacity that Intel does) do more of a subnode solution, where they continuously upgrade the processor subnodes, instead of all of one standard node. The 90 nm represented some changes; they'll have one or two subnodes that enhances the transistor design, or changes certain things in the process technology, or maybe introduces new material, before they go to 65 nm. They're continuously upgrading it.

"So going from 90 nm to 65 nm for them is kind of a slam-dunk," McGregor proclaimed for AMD, "because when they actually make that transition, they've already got most of the solution already up and running."

These aren't architectural or design changes that McGregor is referring to, but manufacturing process changes, some of which may actually be made on the spur of the moment, on the fabrication room floor. Changes are made to such things as gate oxide thickness, or the composition of the doping compound, he said, that come together to improve yields just a bit more each time. "There's two ways to skin a cat," he remarked, "and there's advantages to doing it Intel's way." With multiple fabrication facilities, all running in tandem, there's advantages to being able to formulate a precise process so that all the fabs build products with equivalent performance. "But when you're somebody that's got limited capacity, or maybe only one fab running a particular product, for instance, then you want to make sure you've got maximum flexibility to optimize for yields [and] for performance."

But does this really lead to improvement in product composition and performance on the order of what AMD needs to compete with Intel today? It was process refinement, McGregor said, that enabled AMD to eventually produce Turion processors - in essence, forking off one line of processors from a production process standpoint, and adjusting their operating parameters. Intel would not have been able to make similar adjustments, he said, because there, the manufacturing process is fixed and often immutable. "Intel sticks to one process, and makes everything run on it. AMD optimizes the process for each product. There's a finer granularity of control in manufacturing."

All of which comes down to that big shift AMD will eventually have to make: the change to 65 nm. "You can't just sit there and say [AMD's behind overall] because they're behind in the manufacturing process," proclaimed In-Stat's McGregor. "That's not really true. Bringing up 65 nm for them is a lot easier than it is for Intel, because Intel tries to put all their changes into that transition. AMD doesn't. AMD continuously works in different technologies throughout the process lifecycle so that it's not bringing up a new process with all these changes, all at one time, which inherently is going to create a boatload of problems."