AMD's Phenom: Four Cores on One Die

"After about ten months of watching somebody else marching ahead as the all-around leader in both price and performance, AMD this morning stopped making purely defensive plays, and at last launched its counter-offensive. It will be introducing a new CPU architecture for the second half of this year, aimed at performance-hungry customers perhaps willing to pay a premium.

With the Phenom processor series, which will include a single-die quad-core and a double-quad-core package, AMD will soon be managing three consumer desktop CPU lines, as Athlon moves into the midrange, mainstream space, and Phenom assumes the company's high-performance mantle from Athlon FX.

It is perhaps the last trump card in AMD's hand, and the company may have no choice but to play it now: Since its earliest entry into the multicore space, AMD has used an architecture which moved the memory controller onto the die itself, eliminating the need for a front-side bus architecture, simplifying the chipset, reducing power consumption, and expediting memory transfer through the HyperTransport bus.

Up to now, AMD has delivered so-called "quad core" by way of a dual-socket design for Athlon FX dual-cores. But with the new Phenom architecture in place, AMD can pull four cores into the same die, letting them share a memory controller and an L3 cache while delegating separate L2s and L1s for each core. It's AMD's on-die memory controller design which has been the company's hallmark for the last three years, and it will rely on that design yet again to pull it through.

But as Phenom comes into being, AMD will be a whole manufacturing process generation behind Intel, which came from behind last year in stunning fashion to wrest back not only the price/performance crown but huge chunks of both market share and consumer confidence.

As AMD is just now moving into the 65 nm generation beginning in the second half of this year, Intel is gearing up for 45 nm retooling in the same timeframe, and appears further along in the adoption of breakthrough HK+MG transistor technology - an advance which AMD partner IBM claimed to have discovered on exactly the same day, though which may not be ready for AMD until later next year.

So the challenge before AMD will be to win back the hearts of the enthusiasts, the performance buyers, and the system builders and OEMs, while at the same time keeping their minds from getting too curious about the massive missing performance margin that used to distinguish it from its chief competitor. As BetaNews learned from AMD in recent days, the Phenom campaign this time will be less about statistics and proof points, and more about feeling - about whether the product is marketed well enough for users to feel satisfied, even if it turns out the performance edge over Core 2 Duo and Core 2 Quad is negligible.

"You're going to see more consumer-like marketing come from us than you've traditionally seen in the past," stated Ian McNaughton, Phenom's senior product manager, in an interview with BetaNews. "You're going to hear us talking about our products in a different way than we traditionally have, and it's not going to be, we're sending a data sheet. It's going to be more terms like, 'Exquisitely powerful, intensely visual."'

Indeed, that's the new slogan for Phenom technology that AMD will begin "beta-testing," if you will, today. "When we look at our architecture as it stands today, with four cores, it starts to really shine," McNaughton continued. "The actual benefit of having our architecture versus our competitor's architecture starts to really become apparent."

This is the endgame AMD wants to see, at least for this round: It wants its new Phenom processor line to feel right, to give the appearance of smoothness and sure-footedness, in an everyday work setting or in a hard-driving gaming environment. Maybe it won't win every benchmark - at least not any more, not in the competitive market AMD helped catalyze. But AMD wants it to have that "certain something." And if it tries too hard to quantify it, to measure it, to pronounce it 2% or 12% better than Intel in some obscure contest, it could just lose it anyway.

"When you look at the R&D budgets of the three players in the industry - Intel, nVidia, and AMD - it's unrealistic for anyone to believe that any of the companies are going to be in a leadership position from an absolute performance perspective for a very long period of time," admitted Henri Richard, AMD's senior vice president for sales and marketing, in a webcast a few weeks ago.

"I think that the game has changed when AMD came out with AMD64 architecture, and launched Opteron. Until then, Intel had this almost undisputed monopolistic position. It was challenged in 2000 by Athlon, then they came back with Pentium, and then when we came with AMD64 and the Opteron processor, people thought, 'Well, this is just a repeat of the same game.' And I'm telling you, the game has changed a lot. It changed a lot because we merged with ATI, and so we now have a complete platform."

AMD needs to give its new performance customer the feeling of reliability from Phenom. To accomplish this, it's relying fairly heavily on Microsoft, whose software is probably more responsible for the "experience" factor than the CPU, and on ATI. For the first time, AMD is timing its Phenom launch to coincide with ATI's new Radeon HD 2000 graphics cards, and hoping performance customers will perceive them together as part of a collective platform...and that they won't mind the platform name "FASN8."

"We're bringing emotion to our products," AMD Phenom product manager Ian McNaughton told BetaNews. "When we were AMD traditionally, we've been a processor company. With the acquisition of ATI, now we're a platform company. We're an experience company. And that's something that we're wanting to get across with our processors: It's not just about doing subtraction, multiplication, and division quickly. It's about doing more with your PC. It's about the experience with your PC, and how a quad-core or a Phenom dual-core will help you with that experience."

If you've ever listened intently to a Microsoft presentation, you'll hear a lot about what "it's about" and what "it's not about." It's a very simple method for steering the viewer's mind toward the "takeaway points," and away from the unpleasant facts you don't want the viewer to recall.

Take, for instance, Intel's recent demonstration of the 80-core stack, as an example of the extrapolation of its current architecture. Isn't that a demonstration that Intel has a cohesive roadmap, and is well on its way to executing that plan?

"AMD is not about how many cores," answered McNaughton. "AMD, when you look at our future...the Fusion and Torrenza initiatives that we're doing, it's about heterogeneous cores. It's about having the right technology for the market. It's not about just throwing more cores [into the mix], because that doesn't do everything for everyone...You can throw more cylinders in a car. It doesn't necessarily make it better."

What about Intel's new "cadence," already moving to the 45 nm generation on the heels of just having successfully launched its 65 nm generation? Pay no attention to those nanometers behind the curtain, AMD tells us.

"No consumer buys nanometers," stated McNaughton flatly. "So whether we're at 90 nm or 45 nm or 32 nm, it doesn't make a difference. To the consumer, they're not buying nanometers. They're buying the performance and the experience of their whole platform.

"So what do nanometers do? Let's be frank. What do nanometers give you?" McNaughton went on, challenging us to explain the whole intrigue behind the need for things to get smaller. "It gives you a more efficient manufacturing process, which reduces power, increases yields, and reduces costs. As you move through the manufacturing steps...it's going to be cheaper for you to make those wafers, and you pass that on to your consumers and your OEMs. So it's good to get to 65 and 45 and what not, but now unnaturally you do things to get there, that's really what makes sense, right? Our customers are not banging down our door saying, 'You need to get to 32 nm tomorrow!' Because there's no benefit."

No benefit? What about the fact that smaller processors naturally consume less power - a fact AMD taunted Intel about all through the long summer of 2005? "Why don't you take that and look at current CPUs that are available in the market in the server space, where it really, really matters, and look at performance-per-watt leadership?" McNaughton challenged us. "And we have performance-per-watt leadership. That tells me that either our competitors have an incredibly inefficient architecture or manufacturing process, regardless of what nanometers, or we are very good at what we do."

Then we heard an argument we didn't expect: a contention that there's only so much of this shrinkage a company can do.

"There comes a point in time when there's diminishing returns," McNaughton told BetaNews. "I don't think we're there yet, and everyone's still keeping pace with Moore's Law...[but] I don't think you're going to see a big benefit to [Intel] to move to 45 [nm]."

He explained that Intel may perhaps need the continued die size reduction more than AMD, to compensate for Intel's other design inefficiencies, as well as in order to improve yields - implying they really needed to be improved. "We are focused on delivering the products to our customers that our customers are asking for, and we're committed to giving our customers the innovative products that we've traditionally always delivered."

The message we received from AMD this week was this: Do we really feel all that better about the performance improvements Intel claims to have delivered? Maybe we can see the difference in the benchmarks, but is it something we can truly appreciate? If not, perhaps there's something else that defines quality in the hearts of the performance buyer...if not yet their minds.

Whatever that something else is, AMD may need to get a handle on it quite fast. As SVP Henri Richard told reporters two weeks ago, thanks in part to Intel, the CPU market doesn't work the same way in 2007 and into 2008 as it did in 2005.

"I think what you're going to see is really a change of the game where leadership position in the absolute benchmark that a lot of the enthusiasts are looking at, may change as quickly as every six months," Richard confessed.

"And that's good, frankly. That's good for the industry, that's good for the end user, competition is a great thing. Sure, if we can have leadership for a long period of time, that's something that's a position you enjoy. But I really think that the dynamic of the market has changed, and without wanting to take too much credit for it, I still think we can thank AMD for having changed that dynamic. And I think the end user will benefit from that."