Advocacy

  Myths
  Press

Dojo (HowTo)

  General
  Hack
  Hardware
  Interface
  Software

Reference

  Standards
  People
  Forensics

Markets

  Web

Museum

  CodeNames
  Easter Eggs
  History
  Innovation
  Sightings

News

  Opinion

Other

  Martial Arts
  ITIL
  Thought


Is Intel doomed?
IA32 -- is Intel losing control of x86?

By:David K. Every
©Copyright 1999


Intel was lucky to stumble into their position of dominance in the PC industry. IBM basically chose the worst processor choice available, because they intentionally didn't want their PC's to be too powerful (and threaten the mainframe and minicomputer business).

Since then, the x86 series of processors has gotten a lot faster -- but architecturally it is little better. The PC market wants faster -- but nothing that disrupts the status quo and breaks all the software (and it is too easy to break software in the PC world). This is why Intel has been unable to move from their antiquated x86 instruction set, which has legacy that goes back to 1974, to a newer RISC like instruction set (despite the performance, size, power and heat advantages).

The only thing that has been keeping Intel's head above water (being drowned by the many x86 and RISC competitors with superior designs or implementations) is that they are the leaders. Not better -- just bigger and first. People want to follow someone, and if they are going to follow someone, it is going to be the big-guy. That has worked for a while -- but it looks like the market just changed.

Intel may lose the IA32 market

The problem is that people (other x86 chip makers) aren't following Intel as much any more. People are finally starting to catch on to Intel's behavior and branch out on their own -- and do better.

Intel has a history of changing the x86 constantly in small ways, which forces the cloners to follow suit. By the time the clone-chips makers catch up, Intel is ready to change again. This is similar to Microsoft's strategy of changing their file formats for every new release so that competitors can't make their superior software "file format compatible" (or at least not with the latest Microsoft release).

Changing from Socket-7 to Slot-1 (and all the socket changes up to Socket 7), Slot-2, Slot-M and all those different ways to plug in processors in the PC world, is just Intel's way of trying to break all the clone makers chips, and make them follow Intel. Changing instruction set (slightly) with MMX and now KNI (Katmai's New Instructions), are ways for Intel to try to lock people into their proprietary solutions, and to keep chip-cloners off balance. Intel's monopoly power in the x86 market was because they were exercising control in three areas (the proprietary triad):

  1. Connectors
  2. I/O Chips
  3. Processors.

Intel used their advantage in any area, to help push support for the other two areas. Intel usually changes all three at once -- so if you needed the advantages of any one, you had to adopt the other two. Momentum builds momentum. But changes in each of those markets, may have just eroded the reasons to follow Intel. Lets look at each one individually.

Connectors

The example of proprietary connectors (and Intel's influence), is the socket wars. Intel keeps changing the way the processors are connected to the motherboard. Intel did this so all the competitors would have to spend months (years) redesigning their chips to work with Intel's new connector designs.

Intel helped for the migration, by coming out with new connectors for new versions of their chips, along with new I/O chip sets (2) forcing the PC market to adopt their new connectors -- all or nothing! Monopoly maneuvering defined.

(2) I/O chip sets are what connects the processor to the rest of the computer (and the outside and inside world of the computer). These chipsets are as important to a computers performance as the processor itself. The leaps in both (at the same time) required everyone to follow, or fall behind.

The board makers and system manufacturers adopted Intel's new connector knowing that the rest of the clone-chip makers would catch up (and they'd have a rich supply of alternatives in a year). So they always had other sources -- Intel was just the high-priced alternative for the early adopters. Second sources, and lower cost alternatives would come later.

Of course a year later, when everyone caught up, and a year or so later Intel would have to change sockets/connectors again, since the competitors products were often as good as Intel's processors (or close), but far cheaper. This is why there are 10 different connectors in the PC world.

The problem is that this time it backfired. Slot-1 (what PentiumII's use) was superior to Socket 7 (the last generation, that all the clone makers currently use) but Intel isn't licensing Slot-1 and made it completely proprietary. There is no copying it, and no licensing of it. Intel made it Slot-1 versus everyone else. They were hoping that the performance advantages would put all the clone-chip makers at a disadvantage. It did, for a while -- but a short term success may be a long term failure.

The advantage of Slot-1 is that it supports a backside cache connector (to L2 cache), faster BUS (now up to 100 MHz) and a few other minor things. But people couldn't copy it. So what do they do? They figure out how to make the old design faster, and come up with new designs of their own that are better. If either catches on, Intel will have lost control of sockets, and their little plan will have backfired.

  1. Nothing is preventing Socket-7 running at 100 MHz, or faster. The chip cloners realized this, and have made a super-socket-7 that supports that speed -- with or without Intel.
     
  2. Socket-7 isn't designed to run a backside L2 cache -- but you could do that with a little "interposer" (a board that goes between the chip and the socket that has its own backside L2), or better still, you can just put the L2 cache right on the processor itself (which is even faster still). Some x86 chip cloners are doing that.

    With those two things, the performance advantages of Slot-1 have evaporated, because the clone chip makers had to design around the problem of be locked out of Slot-1. Intel screwed themselves.

  3. Designers can use the Slot-1 connector -- they just have to change the way things are pinned-out. With minor changes the board designers can use cloners Slot-X (connector compatible with slot 1). Some companies are starting this -- if they make it a standard, then many of the chip makers may adopt it, and the non-Intel standard may be more standard than the Intel version.

The big mistake for Intel is that they made it "the rest of the industry versus Intel" one too many times. The rest of the industry is going to rally around something other than Intel, because they have no choice. Intel won't let them copy, so they have to make their own standards -- and they are doing so. That standard will eat up the low-end of the market for sure -- and the PC has been traditionally focused around the low end (who can do it cheaper). If there are no performance losses and it is cheaper, cheaper will win. It is also about momentum, usually Intel's -- but all the clone-chip makers have momentum too... and they just chose a different path than Intel. The costs of jumping to Slot-1 were just too high, and now the performance gains are going to be negligible (or performance losses).

All of these things are happening, and spell the end of the connector wars -- Intel will probably lose control. At best (for Intel) there will be a few different supported connectors, which weakens Intel's control. At worst, Intel will eventually have to come crawling back, and adopt the same connectors as everyone else. While this isn't likely to happen too soon, it seems to already be happening in the low-end of Intel's market (and Intel is considering more socket versions of its processors).

I/O Chips

The second part of "Intel's silicon-triad of control" was the I/O chips (support chips). Intel often came out with the chips at the same time as their processors and sockets, and forced everyone to move at once. Since Intel made the processors, sockets and support chips at the same time, other I/O chip vendors were at a disadvantage -- they couldn't get all the info and support they needed to keep up.

So Intel would beat others to market, and get the leap in mind share/marketshare. I/O Chip cloners would soon catch up, and offer far more diversity in the marketplace than Intel gave (support more types of RAM, more slots, more on board I/O, more processors, and so on). Whatever Intel didn't offer, these guys would fill the gap -- and in many areas they were better. So Intel would change connectors and processors again, and force all the I/O Chip makers to try to catch up again.

Intel tried to starve all the I/O Chip cloners out with Slot-1 as well. The proprietary Slot-1 design locks them out of making chips for Pentium II's -- (or at least made it far harder). So if they can't make chips for Intel processors, these companies have to work closer with the Processor-Cloners. So the processor-cloners, and the older sockets (or new non-Intel ones) will see the advantages, and Intel's processors will not. Not exactly a big win for Intel (long term), and another Intel advantage may erode.

Combine this with a few flops by Intel, like their Orion chip set having bugs, Intel's chip set for Xeon processor had bugs, the BX chip set is very moody with RAM, and lots of little things like that, and you get the idea. Intel has also delayed adding promised features that at least some people want, like FireWire, etcetera. Everyone is learning that they shouldn't blindly trust Intel, and the I/O chip cloners are getting better, cheaper and safer.

AMD is already looking at working with DEC/Compaq to make the K7 and Alpha work with the same I/O bus, and support the same I/O Chips. The Alpha-Bus is superior (performance) to what Intel has today (high end). On the low end, all the chip-cloners are still making Socket-7 and super-socket-7 chips, that may be superior to Intel's I/O Chips. This seems to be the beginning of the end of the I/O Chip leverage Intel had.

Processors

Processors are the cornerstone of Intel's monopoly power. The controlled the processors future, by always being the highest end (best performing). They leapt and others followed -- they had to, Intel ruled (they were faster). The times they are a changin'.

Processors are based on two things

  1. the process, meaning how they are made (how they are fabricated) -- which influences how small the chip is, how much it costs to make, how fast it can run (MHz).
     
  2. the core, meaning how the chip is designed -- which influences how many MHz it can run and how efficient it is at a given speed. This confuses people -- but two chips running at the same MHz can be dramatically different performance.

So lets look at what is going on in these two areas.

The Core - Intel has put so much focus on IA64 (Merced) that they have not focused on IA32 (x86/Pentiums). Traditionally new cores (processor designs) comes out ever two years. The P6-core was out in '95 (called the Pentium Pro) -- the Pentium II and Katmai are still the basic P6, just with little modifications (MMX and MMX2, which don't help the main processing -- just some specialty processing). The next real core change is scheduled for like Q4/2000 (Willamette) -- and that is assuming that Intel hits its schedules (a rare occurrence), 2001 is more likely the release, and later for volume production. Think of sitting for 6 years on the same core, or what is normally 3 generations of change.

Assuming Intel was ahead (in core design) by a generation or so, they had roughly a two year lead (in '95). In '95 and '96 the Pentium II was better than the cloners chips. It took until '97 - '98 for chip cloners to catch up (or get close) -- which is what has happened. Now many are faster MHz for MHz than the Pentium II's. Huh oh. And the competitors next generation cores are about to come out, and they are superior cores to Intel's. The AMD K7 looks like it will be far superior to Intel's P6-core (the core used in the Katmai processor), and the K6-3 is going to be competitive (but probably far cheaper). Intel doesn't have anything serious (core changes) coming down the pipe for at least 2 more years. So on the high end, Intel may be falling behind.

Some may claim that it doesn't matter that the primary core is worse, because Intel has made a few specialty additions like MMX or that they are going to add KNI (Katmai's New Instructions also called MMX2). Well, AMD released 3D-Now in '98, many month before Intel's KNI (MMX2) is due to hit (some time in '99) -- and they do the same thing. 3D-Now is already gaining marketshare, and has many adopters, they are even licensing it to others. KNI is coming out months later, and offers no serious performance advantages. Intel's is actually responding to a x86 cloner instead of leading -- this does not bode well for Intel. So even in the specialty areas Intel is falling behind.

On the low-end the processor-cloners have many integrated machines, with built in I/O, and computer-on-a-chip designs (for portables and low end). They have many cheap designs and can afford to undercut Intel. Intel may be giving up entirely on this market -- and they haven't really done well in it for years.

The Process - in the past, Intel's saving grace was superior manufacturing (process technologies). They could just out produce the clones, leap to new processes sooner, and they could have bigger margins on the same chips at the same price (but they made up for this, by charging customers far more). But guess what -- that has evaporated too.

Intel has to produce in such huge volumes that they can't move to newer processes as fast as the more agile cloners. They aren't going to copper or SOI for years (maybe 2002). PowerPC's (IBM and Motorola) are already making the leap, and some other foundries are soon to follow -- BEFORE Intel. All the clone chip makers have allied with other foundries (places where they manufacture chips) -- and those foundries are getting better at catching up, and may pass Intel -- and they will be making chips for x86-Cloners. Intel's size, and manufacturing requirements (demand) is hurting them -- cloners can afford lower yields, and more experimental processes, because their demands are lower. So even in process, Intel's one real competitive advantage, they may be falling behind. This could be the last nail in the coffin.

Conclusion

Intel is in for a world of hurt in the IA32 arena.

Intel is being squeezed from the bottom, as they have been for a while. They have already been losing the low end, but now they lost their remaining advantages. Intel just can't change things again, and expect the industry to follow -- they locked people out, and now others are going to engineer their own solutions. This forced the cloners to take more initiative and eat some more R&D costs, but it also taught them to make alliances, to work together and that they don't need to follow Intel.

Intel is about to be squeezed from the top as well. That is new for Intel. Those who rule the top of the PC world makes the rules.

Intel is responding by trying to split their high end into two markets -- mainstream high-end (Slot-1 / Katmai), and server high-end (Slot-2 / Xeon). They hope that if they still control the super-high end (server market) that will help them not lose control, and they are hoping that they can leverage advantages in one of those two high-end markets, to help them in the other. But cloners are adapting there as well. AMD is just splitting their markets in two as well.

AMD is poised to eat alive the high-end mainstream market with the K7. AMD has failed to deliver on time in the past, and may miss their goals -- but of late, they've been much better, and seem to be getting everything right. The K6-3 is also a serious competitor. Intel is going to get a lot of pressure in this area.

Intel tried to push up higher with the Slot-2 server Xeon chip. Realistically it offers very moderate performance increases over their mainstream high end chips for about 6 - 8 times the cost. Some bargain. Intel hoped to choke DEC and the Alpha out of the market (they were a serious threat in the high end), and for Intel to take over that profitable market. Instead they pushed DEC into the hands of Compaq, which can play serious hardball with Intel on the high end. In fact, it was that move (trying to starve DEC out) that also pushed many of the Alpha designers into the waiting hands of AMD (and brought about the K7). And the K7 can be brought into the very high end of the server market as well -- and there is a lot of room there (huge margins) for AMD to eat Intel alive there too. More than that, it looks like AMD and Compaq/Alpha are actually cooperating -- you may be able to drop either processor in either system. If others follow, this could be Intel against everyone else.

To make matters worse still, all of Intel's latest moves to make themselves more proprietary succeeded -- but not with the results they may have hoped. It may have driven the cloners together, and there may be more cooperation. Definitely many have selected certain market segments, and will each attack those segments (handhelds, portables, low-cost, etc.). Because they are all specializing, they will each be taking a different chink out of Intel's armor, and Intel just can't keep competing everywhere at once.

I personally wouldn't exactly sell Intel's stock short -- they still have room for growth. But I think that all their competitors are going to be far outgrowing Intel for quite a while. I feel that AMD and Compaq are both safer bets for growth. Even if Intel is in full panic mode (which I believe) I think they will have a long road to recovery -- normally a few years before companies change management directions, and then a few years before they get new processor designs out, and so on. Big companies usually respond slowly and poorly and Intel is very big.

Most of Intel's best maneuvering tactics are gone, and being in the middle of a monopoly investigation neuters them even more -- so they aren't going to use their best tricks. For Intel's future, they had better hope that IA64 isn't the mediocre-flop that I think it will be -- because they are in for serious market erosion in the x86 camp over the next few years.

Intel is pushing hard for IA64 (Merced) as their last ditch effort to crawl out of the IA32 grave that they dug for themselves. However, I think IA64 is going to be a flop for many reasons that I will go into in a separate IA64 article.

Of course the more erosion, confusion, and differing choices in the x86 camp, the better it is for superior processors like Sun-Sparc, HP-PA, Compaq-Alpha and the PowerPC. If Intel is losing control, then the market is going to be more open to considering other alternatives -- and not just x86 alternatives.


Created: 11/14/98
Updated: 11/09/02


Top of page

Top of Section

Home