AMD's CTO: Intel is an old-school company
Monterey (CA) – AMD is going through challenging times these days, with enormous losses on the one side and a new quad-core CPU that needs to re-establish the firm’s competitiveness with Intel on the other. We recently caught up with Phil Hester and got answers to some of our questions about AMD - the competition with Intel, the strategy behind the Fusion processor. Join us for a chat with AMD’s chief technology officer.
AMD has not really been very talkative about its strategy and upcoming products lately. The silence has put the company in an awkward situation, in which Intel’s marketing is running under full steam promoting the 45 nm processor generation while people interested in AMD’s plans could only speculate how well its next-generation processors will compete against Intel. What we have seen in recent weeks, were lots of rumors among journalists and our readers, but also increasingly impatient investors who grilled AMD executives at the firm’s investor meeting.
At a press briefing last week, we used the opportunity to get answers to some of our most burning questions about the current competitive environment, from the man in charge of technology at AMD. We discussed AMD’s current and near future challenges, opportunities and hurdles for the 2009 Fusion processor, the impact of the discussion about 45 nm, and we were able to throw in a few other questions such as AMD’s thoughts on Apple.
Phil Hester, is senior vice president and chief technology officer (CTO) at AMD. He is responsible for setting the architectural and product strategies and plans for AMD’s microprocessor business. According to his biography posted on AMD’s website, he also chairs the AMD Technology Council, ensuring that product development, integration, and process organizations align technology capabilities with product direction.
Before joining AMD in September of 2005, Hester was co-founder and chief executive officer at Newisys and spent 23 years at IBM.
TG Daily: Let’s begin with a general view on the microprocessor industry. Where do you see AMD’s status today?
Phil Hester: We are in a very competitive position. There is a very fierce competition between us and Intel. You have two healthy competitors here; you are going to see companies ping-pong back and forth. You are not going to see one dominate like they have in the past.
The world has become different with the success of our Opteron processor. Going forward, I believe there is a large opportunity for us on the consumer side of graphics. Overall, we feel very good about our manufacturing capabilities and we feel good about our new designs.
TG Daily: AMD is in a difficult time right now and you are facing many competitive challenges. One of the prominent ones appears to be that, according to recent statements from various AMD executives, Intel has been copying AMD. What is your view on this? Is Intel copying AMD?
Phil Hester: More and more so, yes. And that indicates to me that we are doing the right thing. We are going to continue to have a customer centric view of the world. They go wherever they want to go. But we focus on building customer centric products.
TG Daily: So, if you accuse Intel of copying AMD, what exactly did they copy?
Phil Hester: More and more they figured out that Itanium is a ditch. Obviously, they copied our 64-bit extensions. A lot of the work we have done on virtualization they copied. A lot of the work we have done on power efficiency they copied. By doing Torrenza, we forced them to do what is their version of that idea. We don’t know a whole lot about [their version] yet, but, in general, they soon will be copying the idea of co-processors. So, every major platform innovation we came out within the last two years, in one form or the other, they copied.
TG Daily: It appears that the microprocessor industry, especially in the case of AMD and Intel, is in a pattern of switching the lead. For some time, AMD was in the lead in certain markets, now it is Intel and we may see AMD jumping ahead later this year again. Is this rock-solid pattern or do you think that this pattern can be broken and one company can keep the lead for a longer period of time?
Phil Hester: If there is an open competition, there ought to be a horse race back and forth between two strong competitors. That is what is good about us competing in the industry. But if Intel engages in business practices that may not allow fair and open competition, then there is no a serious problem not just for us, but the whole industry.
Read on the next page: The acquisition of ATI, Intel's advantages and disadvantages, and Fusion
TG Daily: The acquisition of ATI can be perceived as AMD trying to catch up to Intel on a platform level. But you can also see it as a move to create something that is different from what Intel can offer. Do you think that ATI’s technology enables AMD to be different and less vulnerable as a result?
Phil Hester: ATI acquisition was a strategic decision. It was not a reaction to any kind of short term platform issues. That being said, most of the Tier 1 OEMs have told us that they want us to continue to offer a choice and not force them into decisions like Intel does: They do not want to have to pick from a set of Intel chips to get the brand. They want a choice of an integrated and tested platform, parts that work from top to bottom. At the same time, they ask us to be working with certain industry partners to make sure that they have a choice. Clearly, they do not want a single source vendor. They want the option of other suppliers.
AMD's new boards and chipsets for 2007 ...
TG Daily: Intel can out-spend and out-resource AMD several times over, in any field you compete in, if they choose to. Assuming that not only your products, but also AMD’s business would be increasingly different from what Intel does – would that decrease AMD’s vulnerability when Intel grows stronger?
Phil Hester: If you are a large and successful company and almost a monopoly in the industry, there will be a lot of inefficiencies in your company. Yes, they have more resources, but the question is: How effectively are those resources used?
In the past, when Intel got those huge margins, they could enjoy some very ineffective operations internally, both in terms of design and manufacturing. We reset the bar in terms of what is an acceptable gross margin on microprocessors. We can work very efficiently within that envelope. That brought a huge structural change for Intel. They may have cash in the bank, but, to a certain extent, they are an old school company, in terms of how they think about the market and the way they design products with competing teams doing the same design. I do not think that is sustainable.
In the long term, we have the right model. It is not a one or two year discussion. This is a five-year model.
TG Daily: Let’s talk about your Fusion processor. I got the impression that this processor is not so much about the integration about graphics capability.
Phil Hester: In the mid-90s, the GPU changed. At that time, the GPU was not just about graphics performance anymore, it became more general purpose. Up to that point, you wrote low-level code to implement the DX interfaces. But suddenly we had enough transistor density that enabled the creation of a more general purpose engine. This change has taken the GPU to the next level. [The GPU] will become more and more general purpose in the future and there will be more and more similarities between what you expect from a general purpose GPU and a specialized engine.
At this point, you start thinking about it in the context of a sustained architecture that is largely driven by applications. If you look at Windows Vista, gaming and a lot of multimedia content, and you wonder what is the most efficient way to process this software, then you see that there are a number of applications that work well in a parallelized manner. It is the right direction to exploit these applications with a general purpose engine.
TG Daily: When you talk about a general purpose engine, what applications do you have in mind?
Phil Hester: There are things like physical modeling and parallel database search. There’s also a whole bunch of applications that previously could not have been imagined to be running on a PC. Face recognition is one example, real time physics modeling another. I am sure we will find other applications, but, if you think about it, it is a little bit a back-to-the-future journey. Look back to the mid-80s, there were Cray super computers that really drove the whole idea of vector processing. With the attack of the killer micros, that went away. That parallel capability is now coming back to standard technology.
TG Daily: We have heard AMD saying that, with Fusion, one-size-fits-all computing will be gone. In fact, the flexibility to combine different numbers of CPU and GPU cores into one processor sounds like there may be many more processor models in the future than we have today. Does buying a processor get much more complicated with Fusion?
Phil Hester: There will be a much broader range of x86 SKUs over time. That does not necessarily mean that there will be more SKUs in the PC space; it means that there will be more SKUs because of a broader range of products.
For example, I can well believe that the next generation of high-end handsets in the cellphone business will use an x86 processor. In some of these [non-PC] markets, [the processor] will be just internal plumbing and not visible to the consumer at all.
Looking at the PC space, we are scaling up. If you are going after new segments, then yes, there will be more SKUs. But I don’t think that you will see a broader range of SKUs in the client space compared to what we have today. Consumers are typically asking for that entry-thing, that midrange thing, or that gaming thing. I do not believe that this will change.
Read on the next page: Market positioning of Fusion and AMD's pricing strategy
TG Daily: At the time you will be introducing Fusion, there will be a transition phase from traditional multicore CPUs to the new type of Fusion processors. One the one side there will be discrete processors with discrete graphics or separate chipsets, and there will be processors with an integrated graphics core. What exactly is the advantage, the selling point of Fusion when the processor hits the market?
Phil Hester: Consumers should not see any functional differences between today’s discrete CPU/GPU solutions and Fusion. What they should see are smaller form factors enabled by the levels of integration we have talked about, they should see better battery life and better power/performance.
One of the things that are very important to us is that the physical integration does not show up at the application level. Applications that run today will continue to run on Fusion the same way. If you go out in time and start to exploit some of these new capabilities, you will start thinking about doing things on your PC that you can’t do today: Physical interaction, architectural modeling – such as stress engineering on bridges - or simulations of chips. With parallel processing capability you can start thinking about doing these things on a local PC and not just in a simulation farm.
TG Daily: This suggests that Fusion will not remain an entry-level mobile part, as AMD has suggested in early press releases. Will we see Fusion moving into other market segments, such as server and desktop?
Phil Hester: Yes.
TG Daily: But in those segments it may not remain an entry level part?
Phil Hester: We have to be careful how we use the phrase “low-end.” It is a mainstream mobile part. But the more we look at the future of mobile and desktop, the only major difference is that the desktop system always has an optional graphics accelerator slot. I believe that the Fusion processor for the mainstream notebook market is probably incredibly well suited for the mainstream desktop market as well. It is packaged differently and there’s different power and cooling, but it’s the same processor.
TG Daily: Then there is the question about price. By integrating graphics into the processor, you essentially eliminate one revenue source you have today. Will consumers have to pay more for a Fusion processor than they have to pay for a processor today?
Phil Hester: Fusion will not be more expensive when you look at the same level of function. Today, you need a two- or three-chipset solution to get [the functionality of Fusion]. When you put that level of function into the CPU and you can provide improved level of power efficiencies and all the other things I talked about, then that chip should have more value than the individual three pieces you have today.
TG Daily: But before you can claim a dramatic value increase behind Fusion, you will need to convince developers to develop applications for it. History tells us that, at least in the client space, 64-bit adoption has been almost non-existent and multi-core is also slow to make its way into the general software space. Exploiting general purpose capability from a GPU seems to be even more challenging for developers. How do you convince developers to develop general purpose GPU applications?
Phil Hester: The first things on the developer’s mind are features. As an example, in next-generation gaming, you want the characters to interact with the environment in a realistic manner. That is a feature that you can actually see in the game. So, this is more of a natural way to marry things - features they want and functions that are available - as opposed to, if you will, asking to directly exploit the hardware capability. In a way, you have to bring them a hardware capability that enables some of the features they care about.
Read on the next page: Motivating developers, Apple and AMD's view of 45 nm manufacturing
TG Daily: If the adoption of code that exploits the capabilities of Fusion is too slow, would AMD considering developing applications for it - to accelerate the move towards a potential killer-application that takes advantage of the horsepower of the GPU?
Phil Hester: In terms of demos, yes. But we are much more into providing tools and enabling the ecosystem. We work, for example, with the financial community to look at software accelerators. The answer is, we do [application development] selectively, but we do not want to do that as a core business.
On the consumer electronics side of things, it is a little different. If you look particularly at the DTV business, they actually like a pretty complete software stack, to the point that it is almost ready for manufacturing.
TG Daily: Since we are talking about the consumer space, there is an IT company that is closer to the consumer electronics industry than pretty much any other IT company: Would Apple be an interesting customer for AMD? Or do you feel that Apple is pretty much a territory that is occupied by Intel now?
Phil Hester: Well, Apple has to answer this question. But, yes, we would welcome any additional x86 customers. There is nothing I know of that is Intel-unique in their code. I would expect that you can run their applications on an AMD processor to the same degree they run them on processors they have today.
TG Daily: Let’s talk about manufacturing. Intel is highlighting their lead in 45 nm manufacturing these days. How important is 45 nm in the industry right now?
Phil Hester: Customers don’t buy nanometers. That being said, there is a reasonable window you need to be in within the industry, from a performance and a density standpoint, to stay competitive. Unfortunately, Intel has tried to portray nanometers as a single metric for competitiveness and that simply is not the case. If that was really true, why is it that Opteron outperformed Intel? [Opteron launched as 130 nm chip in early 2003, while Intel’s competing CPUs were transitioned to 90nm in late 2003 - ed]
TG Daily: But smaller chips result are likely to result in improved production economics …
Phil Hester: You can end up with the worst cost structure in the new technology, if your yields are poor to begin with. Where we really try to focus on is being in that competitive window I mentioned before. Then, we need to transition to a new technology at the optimal economic point.
What we found is that if you don’t push the absolute leading edge of the technology envelope and let the maturity happen for a period of time, the economics turn out to be actually better than pushing it upfront. Yes, you can’t be forever late. But shipping first at a given level of new technology does not necessarily mean best cost.
TG Daily: Thank you for the interview.