Mad Man Sanders - Trash Talking the Pentium Pro - 5 years ago.
Same old story then - same old story now !
"The reality is the design that they came up with is inferior to Pentium in its performance at a given clock speed running a normal mix of instructions.
The P6 is just a goddamn product. "
And yet another Masterful prediction by the Shareholder-Sucking Sanders:
"Sanders: We think the K5 will have a four4-year life, if you count 1995. "
{=================================}
November 3, 1995
Sanders shoots for the body
Interview: AMD chairman plans major desktop thrust in '96
By PC Week Staff
--------------------------------------------------------------------------------
In a bold move to shore up a widening technology gap between its delay-plagued K5 processor and Intel's latest-generation Pentium Pro, Advanced Micro Devices surprised the industry last month by making a bid for rival chip maker NexGen Inc.
AMD's proposal, a stock swap worth about $900 million, should result in the acquisition of the Milpitas, Calif., company and its advanced Nx686 processor design by early next year.
AMD's outspoken chairman and CEO, Jerry Sanders, discussed the significance of the NexGen buyout and its impact on the CPU landscape from AMD's Sunnyvale, Calif., headquarters last week with PC Week's desktop computing editor Steve Kovsky, senior writer Tom Davey, and PC Week/Inside senior editor Bill Snyder.
PC Week: Two weeks ago, you dropped the bombshell of AMD acquiring NexGen. How did that deal come about and what purpose will it serve for AMD?
Sanders: I think the first thing you have to recognize is that in the microprocessor business, that is the Windows-compatible microprocessor business, we're in a marathon, not a sprint. And staying power is what's really going to count here.
Advanced Micro Devices has been able to build a strong presence as the leading alternative over more then a decade. We've produced and shipped, which is even more important -- sold actually, which is even better yet -- more than 80 million x86-compatible microprocessors and we're the only company that's ever produced more than 10 million in a year except, of course, Intel.
In the earlier stages of our career, we were pretty much building Intel clones. Then, of course, we began to build our own re-engineered versions of Intel's 386 and 486, but incorporating the licensed Intel microcode. This enabled us to establish a strong presence in the marketplace and a conviction that the marketplace wanted an alternative solution: that monopolies were bad, competition is good, and the customers wanted to be able to differentiate their products.
So, with that in mind, we recognized that under our strategy of merely being an alternative source, which means you'd always be late to market, we wouldn't give the customers the solution to their problem of having an alternative to an Intel-centric universe.
To that end we began in 1992 our K86 Superscaler project. Now the K86 Superscaler project is probably the most ambitious program that we ever embarked upon, because what it was going to do was give us a competitive alternative with a completely independent technology.
Now the reason I say that it was an extremely aggressive position to take was that what we really want to provide to our customers are solutions, not merely a processor core. However, the processor core is absolutely central to having a solution.
Therefore, our view was, unless we could produce a competitive Windows-compatible core processor, that could be produced in huge volumes so that we could drive material market share, we would not be able to be the alternative solution. To that end, we put our best people, our best and our brightest on the K5 project -- Krypton.
PC Week: How are you defining "solutions?"
Sanders: We're supplying solutions, that is, we want to provide all of the necessary technology to enable our customers -- who are the PC makers -- to offer the feature set to their customers that they want to offer.
Now, that doesn't mean we have to necessarily produce all that technology, but we have to be the nucleating point. That means that we have to become a virtual mother-board provider. By that, I mean where we can provide the processor core and the information to chip set suppliers and other peripheral chip makers, BIOS people, and so forth, so that the customer -- whether it's Compaq, IBM, NEC, Acer, Matsushita, whomever -- [can provide a particular] feature set for the commercial market, the consumer market, or whatever.
Whether that feature set includes audio, video, whatever interactive things, modem, network compatibility, we wanted to make sure that we could offer the complete solution.
Our view was that the solution is built around the microprocessor and the feature set that the customer needs will be satisfied by some partitioning in the system between the microprocessor and the balance of the system.
So what we wanted to be able to do is get some systems expertise and become a high-volume alternative for the microprocessor chip for that core technology. That's the way we would be a viable contender and alternative long-term.
PC Week: Last week at a press conference, when you addressed this question about being a virtual motherboard supplier, you said you had no interest -- unlike the competition -- in being in the motherboard and chip set business.
Sanders: Well, you want to focus on where you're best. AMD first and foremost is a semiconductor IC supplier. We want to be our customer's favorite IC supplier.
First and foremost, what they look for from us is competitive process technology so we can implement the solutions in silicon. Secondly, they look for enough volume so that they can, in fact, get the cost they need and have a viable alternative that can provide them with a substantial portion of their needs.
We'd like to provide, for example, half of the microprocessors used by five of the top six PC makers. Well you run those numbers, and it's says you have to be a pretty big volume supplier.
We'd like to have that not be more than half of our business, so you have to double that, and that gets you up to this neighborhood of this 30 million units a year by 1998, which we think makes us a credible alternative.
Of course, the third thing that the customer wants is a differentiated solution, so that he has some advantage. But it has to be a differentiation within the standard. And that means it has to run Windows applications.
So we put that together, and we say, how do we go about doing that? We want to focus on the microprocessor chip and certain other chips where we believe we have a better idea. But we don't believe that: No. 1, we have all the good ideas in the industry; and No. 2, we're not trying to subsume all of the profit in the value chain.
Our business model is to make acceptable returns -- and acceptable returns means more than a 20 percent return on equity as a company, and we've been consistently doing that over the last five years -- without subsuming all of the profits in the chain.
So to jump back to the question, when you asked me to tell you about what drove the NexGen merger, all of the things I said require, of course, that we have competitive designs.
We believe that the K5 is a better idea then the Pentium. It's a superscaler technology. It uses a unique decouple decoding technique to take the x86 instructions and break them down into simpler RISC instructions and execute four of them in parallel.
That concept is not dissimilar from the concept that NexGen uses in their Nx686. Nor is it dissimilar, frankly, from what Intel uses in their P6. Basically, the concept of a risk platform executing with a high degree of parallelism in superscaler fashion these reduced instruction sets, these RISC instructions, is the better idea.
After you get the better idea, the name of the game is execution. So we put together this young team of people in Austin, Texas.
Well, we burned the boats. There was no going back, no clone strategy. We were going to have an independent solution in the K5. Now, the K5 remains a better idea, we're just late, based on our expectation.
This being late with our expectation has caused me great grief in the press, hammered my stock price, and caused me great personal anguish.
The facts are, these guys were pushing the envelope, and if in fact, we have the product as advertised in the early part of 1996, it will be less then four years from conception. Intel started their P6 in 1990, so if they have their part here, they're talking five years.
The realities are that it was unrealistic for us to expect that we could have this kind of a breakthrough in a three-year period when fine competitors were taking four years.
The truth is Intel may be introducing microprocessors generations on a shorter interval, but it's only because they started them a long time ago. We're in a marathon, not a sprint. The only product we started in 1992 was the K5. We didn't start anything else's in 1993. By the time we got to 1994, we were scratching our heads on K6, and of course, now we're beyond the head scratching stage on K7.
Without some additional resources brought to bear, AMD would have had a gap in its product line after the K5. The opportunity to acquire NexGen presents us with a 6th-generation processor in the time frame of interest.
So the driving issue in the acquisition is the NexGen Nx686 -- once we learned it had a P54C-compatible pin-out and bus.
Our thrust is the desktop. We shoot for the body. We don't want to settle for "winging the enemy," and we certainly don't want to shoot him between the eyes and miss. Now we'll be in a position to be participants in the mass market in a big way. . . by the middle of 1996.
PC Week: With the K5?
Sanders: With the K5. And in 1997, we will be able to also be a major player in the performance desktop [with the Nx686].
So the way I see it, the NextGen combination gives us the wherewithal to achieve our objective. Of course we have to perform, but the opportunity is there to achieve our objective of having more than 30 million processors shipped in 1998, which we believe will be a 30 percent unit market share. And it should get us well on the way to having more than half of our $5 billion in revenue in 1998 be from x86 processors.
PC Week: Do you see future generations as being based on NexGen's architecture or AMD's?
Sanders: Each generation actually is a unique architecture, by nature of the way you put the thing together. But I see the 7th generation as being a beautiful marriage of the best ideas of both companies.
PC Week: There's some possibility that Intel's 7th generation will be a VLIW (very long instruction word) chip. Is there some possibility that your 7th generation chip will incorporate some idea like that?
Sanders: The answer to the question will not be coming from this mouth today. We're staying with our program of shooting for the mainstream market. We don't believe that it's necessary to do -- strike that. We want to supply a $100 microprocessor.
The Intel model is to have a much higher average selling price than a $100 microprocessor. In order to make that model work, they have to subsume the rest of the system and they can thereby maintain their business model, but at the expense of their customers.
PC Week: Does that perpetuate AMD in the late-to-market role, if you're always playing cleanup with chips that compete with the low-end of Intel's line?
Sanders: No, I'm not proposing we play cleanup.
We are not trying to be the Mercedes Benz 600 of the microprocessor business. Small companies think that they're going to beat Intel by having some design which allows them to cherry pick the market in selling these $300, $400, $500, $600 processors. Lots of luck, guys.
PC Week: Do you have a target average selling price in mind for the PC. There's been talk about the $1000 PC, some people even talk about a $500 PC. What do you talk about?
Sanders: I talk about a PC where we will offer chips which can run Windows with 486DX4 100MHz performance at a price of $25 before the end of 1996. I think there's going to be opportunities for guys to build what I will call "not full-featured PCs," which may be able to do E-mail or be some sort of a network browser. Maybe that's a $300 box. You can't build that with a $400 microprocessor in it.
PC Week: Do you have any plans for a low-end multimedia chip, like Cyrix's recent introduction?
Sanders: We intend to have the K6 have multimedia capability on it. When NexGen announced the 686, they pointed out that it would have an MEU (multimedia enhancement unit)on it. We are reviewing our K5 product plan to consider whether or not we add such a capability in the 5th generation.
PC Week: NexGen's design must use the IBM cell library, it must be optimized for IBM's tools and processors, since IBM is their fabrication partner. How difficult is it going to be to port that design over to your library, to your tools and processors?
Sanders: Well, it turns out, it's not a slam dunk. But I merely point out, that is what we reviewed during the months of the merger discussions.
We have had in-depth analysis of what's required, and our people are highly confident that they can make the appropriate changes so that we can run the NexGen design on the AMD process consistent with having first units available in the second half of next year from AMD's Fab 25, and have volume production in 1997. So, I feel comfortable with that program.
PC Week: It's no trivial task to change the process. Of course, one of the important lessons you learned in the development of the K5 is that this is truly "rocket science," and some things take longer than you expect. You don't think that porting over from the IBM process might take a little longer than you predict?
Sanders: First of all, I'd like to say there are two very different things here. The problems we've had with the K5 were the validation of its compatibility. That is the ability to take the instruction set, decode it, and then reassemble it and have it work with 25 to 30 different operating systems, including network operating systems, and an infinity of applications.
This is rocket science. You are right. We have a whole different approach to taking the NexGen design rules that will be required for the AMD process, working with the NexGen designers, the AMD process people.
This is a different task. Sure, it's not without its difficulties, and a few things have to be modified to ensure that we comply with all the appropriate restrictions on intellectual property. We feel that we can do that. This is not the same task as the compatibility.
We've done this before. I remind you when Intel gave us their first database tapes for the 8086. We had the thing in production in less than 90 days. And we had a different process. We modified the process and the database tapes very quickly.
AMD is a process technology leader. By the way, the integrated circuit business is a process technology business.
Increasingly, it's becoming clear that now all of the excess capacity in the industry which made the concept of a fabless semiconductor company, an IC company, viable is fast going away.
Now you need to have your own fab capability. Moreover the coupling between the design and the process allows you to optimize performance.
There will be the first K6 and then there will be improved versions of the K6 as we marry these technologies together. Because, remember, we want to put this thing on .25 micron as soon as we can. It's going to be on .35 [initially], and when we go to .25 micron, the NexGen design will be optimized for the AMD .25 micron process.
PC Week: When will that be?
Sanders: I'd rather not say at this point in time, but I've committed us to having .25 micron capability by the end of '97.
PC Week: AMD has a tremendous amount of very high quality capacity, yet NexGen officials said last week that the relationship with IBM Microelectronics is going to continue and he wasn't just talking about just fulfilling contractual obligations. Why is that? Can you foresee IBM manufacturing for AMD?
Sanders: The important issue is having enough volume to be relevant.
The concept of having IBM as a manufacturer of an alternative solution to an Intel product is worth pursuing. So we are open to the concept of further collaboration with IBM on a cooperative basis.
IBM is one of our largest customers; we provide IBM PC Co. with 486s; we have a high level of interest in our fifth generation, our K5 generation from them and others; and from a sixth generation point of view, clearly we are open to a relationship on an ongoing basis.
PC Week: Are you thinking that you don't have the capacity to get to that 30 million unit target?
Sanders: I have the capacity, depending upon the die size. We have programs to drive to the point where we can meet those capacity numbers.
But remember, I said 30 percent unit share. That presumes we're going to be on the entry level, in the mainstream market. I'd like to see an alternative solution command more than the current level of non-Intel dollars.
This week's Business Week showed 85 percent of the x86 revenues were Intel. So AMD has never had more than a 20 percent unit share. We've gotten up in the 17 to 19 percent range, but clearly we've never hit a 10 percent revenue share.
We want to get more revenue share for the alternative. If that means sharing some of the business opportunity on a global basis, we're prepared to do that.
PC Week: Jerry, you mentioned IBM has a high level of interest in the K-5. Have you got any new significant OEMs for K5 and also for the NexGen chip? Will they add any new significant OEMs to the stable?
Sanders: First of all, let me comment on NexGen by saying I can't comment on NexGen. We don't own NexGen. These are two separate companies, and we must operate them separately.
What we're doing of course is having discussions to make sure that when we become one company, we don't lose any time.
Now relative to other new OEMs for K5, I'll just remind you, there's no point in having an OEM for K5 until you've got a K5. The level of interest on a global basis is enormous, absolutely enormous, and we will be sampling a K5 to demonstrate compatibility by year's end.
Now this has been the major issue for us. The first issue is demonstrating compatibility. We feel that we are now to the point that the product, which we just taped out and which we'll have available for wide-spread sampling, will have all the compatibility required for people to design it in. Then we can begin to engage with OEMs with a view towards actually securing business and commitment.
At this point in time, we're not engaging with any OEMs until we can demonstrate that compatibility.
Now the first product that comes out is an entry-level product. It'll run at 75MHz and it's a very attractive part, but strictly as an alternative. Then we will proceed with the higher-speed versions with a view towards driving some market share in '96.
PC Week: What kind of lifespan will the K5 have?
Sanders: We think the K5 will have a four4-year life, if you count 1995.
PC Week: We're hearing that Intel was counting on the Pentium Pro being a high-volume, mass-market chip at the time of its introduction, but Microsoft wasn't able to execute with making Win 95 a completely 32-bit platform, and we have the scenario we find today. What's your take on it?
Sanders: All Fantasyland. Intel rewrites the history to serve its purpose.
When they came out with the 386SX, which was supposed to be the 286 killer -- the AMD killer -- they said this SX was a simplified DX. It didn't have a floating point processor, so that's why they called it the 386SX. Then when the 486 came along and they wanted to kill the 386, because AMD had now stolen the market for 386s, they came out with a 486SX. They didn't take off the processor, they just disabled it. And they said the SX means simplified?
Intel continues to rewrite history to serve their purpose. It's pretty clear to me that in 1990, when Intel was developing their sixth-generation processor, they weren't sure what they were going to do with it. They had an idea, similar to K5, that is, to take the x86 instruction set, run it on a RISC platform, and get blazing performance.
They just didn't think through in 1990 [when Intel started development of the Pentium Pro, or P6] what ratio of 16-bit there would be versus 32-bit instructions [in 1995]. The rest is just a made-for-television story, created after the fact.
The reality is the design that they came up with is inferior to Pentium in its performance at a given clock speed running a normal mix of instructions.
So what does that mean? Well, let me be the first one to say it: The guys who conceived of the Pentium were better than the guys that conceived of the P6.
The Pentium is a better part, and it's got lots of legs. They're talking about 180MHz parts, maybe going to 200. The Pentium is a goddamn good product. The P6 is just a goddamn product.
PC Week: You just hired one of the guys [Vinod Dham of NexGen and formerly of Intel] who conceived of the Pentium.
Sanders: I didn't know that in advance. God smiles on me. Character pays.
There's one more thing I want you to understand about this P6.
The P6 is targeted at the server market, and it's a fine server product. It goes a long way when using optimized compilers and when you're running 32-bit code. It's a very powerful processor, and I think it's going to make a fine server. And they're going to be very successful with that product.
That's not our market, that's the server market.
Intel has pushed out a battery-powered P6 until the first of 1998. What does that tell you? It tells you that this product was not optimized for the desktop and then to be converted into a high-volume portable.
It says somebody back in 1990 made a call to build a powerful processor with a new idea of RISC-like performance, and now what you see is what you get.
There will be other P6s, and they won't look a lot like the P6 that currently exists. There'll be a lot of changes.
But Intel controls the press. They control you. So they tell you some story that you know that some guy forgot that 16-bit would still be around and that's why it's a piece of crap running 16-bit. Well, maybe. And maybe it just wasn't a very good solution to the mainstream desktop. That's where we see it.
PC Week: So, what is Intel going to do in the interim?
Sanders: Sell the hell out of Pentium.
PC Week: How far can they ramp Pentium. How fast can they push it in terms of clock speed?
Sanders: It turns out the customers don't even need it now. The facts are that a 100MHz Pentium is more than adequate for almost anything that's out there. Christ, with 133MHz or 166MHz Pentium and some multimedia extensions you can probably do MPEG-2.
So what do you need beyond a 166MHz Pentium? The answer is a 180MHz K6. We'll have a solution for any task for the desktop.
I think Intel is going to be tremendously successful with the Pentium over the next three years, and I think that that's why they took the brand equity and called this other product Pentium Pro. To take advantage of the mass merchandising muscle, as well as to recognize that the Pentium's got legs.
--------------------------------------------------------------------------------
Copyright (c) 1995 Ziff-Davis Publishing Company. All rights reserved. Reproduction in whole or in part in any form or medium without express written permission of Ziff-Davis Publishing Company is prohibited. PC Week and the PC Week logo are trademarks of Ziff-Davis Publishing Company. PC Week Online and the PC Week Online logo are trademarks of Ziff-Davis Publishing Company. |
