CNET también está disponible en español.

Ir a español

Don't show this again

Tech Industry

AMD's Opteron aspirations

Dirk Meyer helped rescue Advanced Micro Devices from a tight spot once before. With a major chip release only weeks away, the company is hoping he can do the trick again.

Dirk Meyer helped rescue Advanced Micro Devices from a tight spot once before. Now the company is hoping he can do the trick again.

Meyer, who is senior vice president of AMD's Computation Products Group, joined the company in 1996 to become director of engineering for the K-7. The K-7 ultimately became the Athlon, the microprocessor that pulled the company out of a financial tailspin and earned industrywide respect for AMD's design capabilities.

AMD again finds itself facing a familiar scenario. The company is a couple of weeks away from introducing a new chip for servers--this time it's the Opteron, a product that analysts and computer executives believe could establish the microprocessor maker as a major player in services. But at the same time, AMD is swimming in red ink, and the company's comeback depends on the success of the new microprocessor.

Meyer, who was the co-architect of the Alpha 21064 and 21264 microprocessors at Digital and worked at Intel, spoke with CNET about lessons learned from the past, the upcoming chip and AMD's ongoing competition with Intel.

Q: Opteron comes out in a few weeks. What's left?
A: We have a launch event on April 22. That's about it. There are essentially no issues on the manufacturing front.

There seems to be two general technological thrusts behind Opteron. The chip can run 32- and 64-bit software, but it also reduces memory latency. It seems like a lot of the bang for the buck, at least initially, will come in reduced latency.
If you look at the product features of Opteron and compare it with what's in the market today, there are generally three new features. One is the 64-bit extension to x86. No. 2 is the integration of the memory controller onto the microprocessor silicon, which of course is a good thing for memory latency, and hence a very good thing for performance. Memory latency is a barrier to better system performance.

The third element is the creation of

What people want to know is, Can AMD make the chip? and Can AMD deliver the performance as measured by 32-bit applications?
How will the chip roll out? Will smaller manufacturers and white-box guys sell Opteron servers first?
Yeah. The launch event is going to be a big event and when you see the level of participation at the event you will be impressed, but it is just another step along a road toward Opterons everywhere.

What are some of the factors that could tip acceptance for Opteron? Will it be good benchmark tests? Will it take a few major Fortune 500 types to toggle over?
If you stand back, what we are really talking about is the creation or formulation of an Opteron ecosystem and there are a whole bunch of players and agents that are going to need to be part of it. Ultimately, Fortune 500 customers are going to be one contingent. Software developers with 64-bit apps are going to be another. The good news is that the economic value equation is real obvious.

What people want to know is if AMD can make the chip and if AMD can deliver the performance as measured by 32-bit applications, because that is what is available today. When the answer is yes to that, the next question is, What about the 64-bit ecosystem? How far down the road is that?

The real analogy is a snowball rolling down a hill. I can't point to any great event that makes a snowball go from a snowflake to an avalanche, but as time goes by the snowball gets bigger and bigger and rolls faster and faster.

Is the release of 64-bit software going to be more of a 2004 event or a late 2003 sort of thing?
As you know, we've been working with the Linux community for years now. We've got 64-bit kernels running. We've got a 64-bit development environment going. That will essentially be available at launch. What I can't speak to yet is what Red Hat's plans are versus what a SuSE is doing. In general, speaking of Linux, over the course of 2003, you're going to see an increased level of support, and certainly by the end of 2003, all the pieces will be there.

How about Microsoft?
As you know, last April we signed this collaboration agreement. And as every day goes by, I can tell you--and this sounds corny--but I can tell you that I have been happy with Microsoft on this front.

Is Intel working on a 32/64-bit chip? Every time the topic comes up, they stand behind Itanium.
To me, it is not a matter of if, but when. I just think the direction is so clear, and what the customers want is so clear, that given the existence of competition in the marketplace, Intel will ultimately have no choice but to respond. If we didn't exist, I think Intel would be in a position to drive Itanium down everyone's throat. But we do exist, and we are going to offer an alternative and even a virtual monopolist like Intel is going to respond at some point.

Itanium in some ways could be called a historical anomaly. The idea was conceived when Intel was barely in the server arena. You guys weren't looking at the server area. Then Itanium got delayed and delayed. Then Xeon came along.
I was back at Digital when Intel came out with their Itanium plans. It was back in the early '90s. Since then a couple of things have changed. One is that they've got more meaningful competition in the marketplace in the form of us.

No.2, the economics that drive the server business have changed in the last ten years. Intel has been incredibly successful in driving PC economics into the server space and driving x86 into the server space and, in essence, created by themselves a demand for x86-64 in the server market.

We are going to offer an alternative, and even a virtual monopolist like Intel is going to respond at some point.
Speaking of Digital Equipment, it was an interesting company. A lot of ideas for chip design--multithreading, integrating the memory controller--seem to have come out of there. What was going on? Can you trace the heritage of Opteron back to the company?
I don't want to trace specific products from AMD back to Digital, but Digital just had an unbelievably talented engineering team in all sorts of areas. Just look around the industry and look at the people that came from Digital and what they've done. For example, Intel StrongArm was created by a bunch of guys I worked with on Alpha. It was an incredible environment.

Didn't Hypertransport originally emerge out of Alpha?
I think what you are thinking of is that the first Athlon incorporated the EV6 front side bus. Hypertransport was largely an AMD invention.

I thought it came from an Alpha group.
Alpha Processors was a wholly owned subsidiary of Samsung. API was cast with creating chipsets and hardware in an attempt to make the Alpha market bigger than Digital. As a result of them building chipsets using the EV6 bus and us being on the EV6 bus at the time, we had reasons to visit API. We were pretty far down the road on Hypertransport, but we were culturally quite compatible. Some of those engineers contributed materially to the Hypertransport spec. Samsung ultimately closed down API and we hired all of the engineers. Our Boston design center is made up of API folks.

But how do you prevent what happened to Digital from happening to AMD? They had a lot of talent and great ideas, but they could never get past 4 percent to 5 percent market share.
Digital was in a terrible trap. They grew up in the context of the old vertically integrated computer companies with an economic model set up around selling expensive minicomputers with high margins to a captive customer base. As we all know, the computer industry got transformed from being an industry made up of vertically integrated companies to a horizontally integrated industry. Digital could never break out of that trap. And despite the fact that Alpha was a wonderful technology, they were never successful at making Alpha any bigger than Digital's market. This led to the demise of the business model.

Interestingly, if you look at Sun, they were much more successful at getting Sparc big enough to base a business on. But I think that many people now question Sun's business model.

On another subject, the desktop version of the chip, Athlon 64, won't come out now until September. Why the delay?
It was a bunch of things. We wanted to focus our energies on Opteron because that is an opportunity for us to bust into a new market. The

Invite Michael Kanellos into your in-box
Senior department editor Michael Kanellos scrutinizes the hardware industry in a weekly column that ranges from chips to servers and other critical business systems. Enterprise Hardware every Wednesday.

other dimension is that the capability of Barton (a new version of the Athlon chip that came out in February) exceeded, a little bit, our expectations. We're able to sell Barton at the top of the desktop marketplace. Given that we can participate at all of the price points with Barton, there is not a burning need in the short term to come out with Athlon 64. So it made good business sense to focus our resources on Opteron.

Conceptually, Athlon 64 will come out at the top of the market and waterfall down, kind of at the rate our processors normally fall down. You can go back and look at our road maps and draw Athlon 64 along the same chart.

But it will now come out the same time as Prescott, Intel's next major chip. For years, though, Hector (Ruiz, CEO of AMD) and Jerry Sanders (AMD's founder) have talked about AMD's die-size advantage and how a smaller chip leads to lower costs. Prescott will be a 90-nanometer chip. Athlon 64 comes out on 130 nanometers. As a result, does the die-size advantage disappear?
It is probably kind of misleading to compare a 130 to a 90. I guess what I will say is this: At like technologies, 130 to 130, and 90 to 90, we have a die-size advantage. I expect that to be true going forward.

It has also typically been the case that Intel introduces a new technology--be it 180 or 130 or 90--a little bit ahead of us. But we follow very quickly and ultimately transition our factory much more quickly. We transitioned from 180 to 130 over the course of two quarters. It is a difficult question to answer in the middle of a technology transition.

So there might be a little more parity in terms of size between the two chips at the beginning, but we're also talking about small volumes?

Hector also recently talked about how AMD engineers and developers are moving to upstate New York to work at IBM on developing future microprocessor technology? How much of the engineering staff is up at IBM right now?
Our Semiconductor Development Center in Sunnyvale (Calif.) used to house both memory that is flash, and logic technology development. SDC is transitioning this quarter to be devoted to flash. Meanwhile, our logic technology development is going to be done jointly with IBM, and we will have a number of AMD engineers working in East Fishkill (N.Y.). Ultimately, that technology will be deployed in our factories.

So it will be the logic development guys, not the processor designers, who are going to New York?
It is purely manufacturing process development.