the cpu makers

Intel have a massive market share and a long, proud history. Intel have as good a claim to having made the very first CPU as anyone. (Digital are the other firm sharing this honour.) Intel were certainly the most outstandingly successful of all the early CPU makers, and from 1971 to 1995 never once failed to follow a good product with an even better product. Intel's profitability is legendary, and the company continues to grow, though more slowly than it did in the '80s and '90s.

In addition to CPUs, Intel make superb networking products. The compsny used to make a lot of RAM, and now makes a substantial proportion of the world's motherboard chipsets.

Intel's great strength has always been in manufacturing. Intel has considerable marketing prowess, but it is the firm's superb production facilities that have made it into the greatest name in computer hardware.

Curiously, from a design point of view, most of the famous Intel CPUs were not particularly advanced. The 8080 was soon outclassed, the 8086 was slow and clumsy, the 286 a dreadful kludge, the Pentium a triumph of muscle over mind. Even the wonderful 386 suffered from its "brain damaged" parentage. Any veteran programmer knows that the 6802 and the Zilog Z-80 blew the 8080 out of the water, that the 68000 and the TI 9900 were far superior to the 8086, and that both the AMD K5 and the Cyrix 6x86 were better designs than the Pentium or the Pentium MMX.

Intel made up for these uninspired designs — more than made up for them — with unparalled manufacturing expertise. By the time AMD had a 486DX-40, Intel were selling a 486DX-66. When AMD got to DX/2, Intel had a DX/4, and so on. Intel took uninspired designs and through sheer manufacturing expertise make them go consistently faster than anybody else.

As a representative example, think of the three common 166MHz class chips that made up most of the market in 1997, all of roughly equal performance. The Pentium clocked at 166MHz, the 6x86 managed to do about the same amount of work at only 133MHz, and the AMD K5 ran at a mere 117MHz. There are complications to consider, but essentially the AMD and Cyrix chips got their performance by advanced design, the Intel parts by advanced production methods. This is Intel's standard recipe for high performance: very high-tech manufacturing facilities pushing existing designs up to impossibly high clock speeds, but making them work reliably anyway.

In the early days (the late '70s and the most of the '80s) Intel's chips were very run-of-the-mill. They weren't particularly well-designed and they were not fast either. What they were — and this was absolutely vital at the time — was available. Intel CPUs were reasonably cheap, in volume production with good second-sources, similar enough to existing designs to be easy to build a board around, and above all, available right now.

This is why the 8088 was such a success. The 68000 was expensive, the Z-8000 was always going to be ready "next month", and you practically had to get down on your hands and knees and beg before Texas Instruments would sell you a TI9900. Intel's 8086 and 8088 were slow and clumsy by comparison, but system designers could buy them, buy in or make cheap, reliable support chips for them, and get their computer onto the market this year. In a world where start-up companies need immediate cash-flow just to keep on going, time to market is everything.

Intel's magnificent 386 was their crowning glory. In many ways, the 486, the Pentium, and even the Pentium-III are just big, fast 386 chips. It's the 386 architecture that makes Windows XP, Linux and OS/2 possible, the 386 instruction set that powers your Athlon or your Pentium 4. The 386 was a massively complex chip, far bigger than anything which had gone before, and only Intel could have hoped to build it back then.

It had a very American design philosophy, the 386: throw in everything but the kitchen sink, keep stirring until it almost makes sense, then pour in bucket loads of money and transistors until, somehow, it all seems to work. Of course, that makes it very big and hot and expensive, so throw in still more money building brand-new fab plants churning out ever-thinner wafers in ever-greater numbers until the whole thing seems mere routine and you can make them for seven dollars each. Seeing as this is all leading edge technology, and vastly more powerful than the dreadful old thing you were selling last week, you can sell your seven-dollar chip for four or five hundred — and you'll need to, because the back-room boys are just about to do it all over again.

That's Intel.

They didn't quite know how to improve on the 386, so they made it four times bigger, less than twice as fast, and ten times more expensive. This was the 486. Anyone else would have gone broke at this point, but not Intel. They sold the first few slow, early 486s to people who didn't care about the money, the next trickle went to performance-conscious buyers with big budgets, and by the time AMD were killing them in the 386 market, they'd worked another miracle on the production line and the 486 was (almost) affordable. The Pentium was the same thing all over again: three times bigger, eight times more expensive, but only 50% faster this time. The game was getting harder. There are physical limits to how far you can take extra silicon, and the return on investment gets smaller and smaller as the chip gets bigger.

With the Pentium Pro, Intel hit the wall. It was twice the size of a Pentium and for most tasks actually slower. For the first time since the ho-hum 8085, Intel's newest chip didn't set the world on fire. These days they say they always intended it for the server market, where it did well against the SPARC, the PowerPC, the Alpha, and the other RISC chips, but it was really supposed to find its way onto our desktops the same way the 386, 486 and Pentium did. If Microsoft had made Windows 95 a true 32-bit operating system, the Pro might have been just the thing, but no-one really wanted a Pro unless they were running NT, Unix or OS/2, and for Windows 95 users a 6x86 was a fifth the price and noticeably faster. (This is one of the reasons why Intel and Microsoft don't get on very well anymore - Intel have never forgiven Microsoft for telling them that Window 95 was going to be fully 32-bit but then delivering a largely 16-bit product that ran poorly on the 32-bit optimised Pentium Pro.)

Any other company would have been in trouble after a failure like that, but Intel put their bravest face on things and soldiered on with ever-faster versions of the venerable Pentium, and tweaked it up with MMX versions to give them breathing space until the Pentium II arrived. It's a testament to the strength of the company that they could take a blow like this and not just survive but actually prosper. The Pentium remained the biggest selling CPU in the world right up until the end of its market life. Although the Pentium II is mostly just a re-thought Pentium Pro, it soon became much faster. Once more, Intel's manufacturing miracle-workers were responsible: by simply making the transistors even smaller and doubling or tripling the clock speed, the Pentium II once more give Intel the world's fastest X86 CPU.

In the meantime, Intel expanded into other parts of the PC industry, notably into motherboard chipsets and even complete systems. In 286 and 486 days, Intel chipsets were pretty ordinary. But with the introduction of the Pentium, several things happened: Intel brought out some really good chipsets, the then-market leaders, OPTi in particular, brought out some lack-lustre ones, and Intel began to use their marketing muscle and their massive financial resources to wipe out their competitors, one by one. There are many stories whispered around the industry of dirty tricks during this period. Occasionally they make it into print, usually in the trade press, almost never in the mass-market magazines where the big advertising dollars are. Right now (February 1999) it looks as though they are starting to surface in the courtroom. But whatever the reason, during the mid-Nineties Intel became the world's largest supplier of chipsets as well as CPUs.

It's interesting to compare Intel with the other young giant of the PC industry, Microsoft. Both firms grew from small beginnings to almost total market dominance in just 20 years or so. Both have developed a reputation for very tough business methods of dubious legality and little morality. And between them, they have cut the grand old giant of the industry down to size. Sure, IBM is still easily the biggest and most successful computer company of them all, but it's Intel and Microsoft who garner all the glory and, despite IBM's role as the research engine of the whole industry, it's Intel and Microsoft who make nearly all the decisions about our future.

And yet they are very different firms, Intel and Microsoft; the comparison should not be taken too far. Microsoft have had serious competition up until quite recently: from Lotus, Word Perfect, Novel, IBM, Digital Research, Apple and Netscape, among others. Since 8086 days Intel have had hardly any competition at all. Even the AMD 386DX-40 and 486DX/4 were only minor annoyances, and it wasn't until the Cyrix 6x86 and the AMD K6 emerged that Intel had to become seriously worried — so much so that they had to forgo two billion dollars of 1998 revenue with massive price cuts forced on them by Cyrix and AMD. In contrast, it is Microsoft's competitors, Corel/Word Perfect, Lotus and Netscape, who had to make the price cuts, while Microsoft's slice of the average PC pie doubled and redoubled. And while Intel are locked into massive research and manufacturing costs, Microsoft no longer have to bother spending too much money on product development: it no longer matters if Word is as good as Word Perfect, because Microsoft have reached total domination. Microsoft don't have to bother trying anymore, while Intel need to try harder than ever.

Finally, they are companies with rather different slants on business ethics. Both have got bad names for crushing their competitors with unfair, barely-legal trading methods: Intel with their infamous secret black list, Microsoft with a host of (alleged) dirty tricks, big and small, subtle and blatant. (Both companies, of course, claim they are squeaky clean. You make up your own mind.) But there is a difference in degree. For all their strong-arm marketing, Intel have almost never failed to deliver a quality product, and on the very rare occasions where an Intel product hasn't been up to scratch (remember the Pentium floating-point bug?), they have spent as much as necessary to put matters right. Microsoft, by contrast, won't even take your phone calls unless you give them a credit card number, and their standard response to bug reports is to insist that you need to buy another upgrade. (By the way, there is a proposed new law in the USA to make it illegal to sell a software upgrade before fixing the bugs in the existing version - what a wonderful idea!)

The really sad thing is that while Intel's future is reasonably secure, Microsoft's is positively rosy. It seems that in this fast-food television world, nice guys loose. If we had Microsoft's ethics, we'd have been long since out of business in a small town like Ballarat, but on the world stage the rules are different, it seems. Failing miracles, within the next ten years we'll all have lost what little choice we still have even now. Still, I'll be retired by then, with more interest in my roses than my 9x86, so maybe I'll let younger people worry about it.

Intel have always faced some sort of competition. They haven't always doninated the market, after all - in the Seventies and early Eighties they had to claw their way to the top against stiff opposition from the likes of Zilog and Motorola. It was only in the second half of the Eighties that they came to hold a virtual monopoly. In the first half of the Ninties, they fought long and hard to maintain the monopoly, to drive all other contenders out of business, and only a handful of the most determined stayed the course - notably AMD and Cyrix. IBM too were still in the game - though more through sheer size and inertia than determination.

But the second half of the Nineties saw a far more serious challenge. For the first time in ten years or more, competing firms produced CPUs that were as advanced or more advanced than Intel's own - notably the 6x86 Classic (Cyrix, 1995) and the K6 Classic (AMD, 1997). On its own, this mattered little; the overwhelming majority of buyers stuck with Intel and, although it forced some price reductions here and there, and at least one significant product improvement (the Pentium MMX), the net effect of competition was small.

It wasn't until mid-1998 that the monopoly finally threatened to come apart at the seams. There was no single cause for this, it came about as the result of several factors:

1. AMD, always the most serious Intel competitor, produced an outstanding new product: the K6-2.
2. By and large, the market finally stopped believing scare stories about non-Intel CPUs. The cry of "wolf" had gone up once too often.
3. Intel's product development teams produced a series of lack-luster new parts.
4. Intel's product planing people comprehensively failed to predict the massive shift in the market towards more affordable systems during 1997 and 1998. (AMD and Cyrix were in a much better position to benefit from the market shift. Maybe they planned it that way - but we suspect that they just got lucky for once.)

Let's look at Intel's product development in more detail. On the whole, the Pentium Pro had been a flop, and the Pentium II was very slow to establish a significant share of the mainstream market. Unlike AMD and Cyrix, Intel missed the opportunity to develop a relatively small, modern chip a with good price-performance ratio. The Pentium design was simply too old and too slow to stand up against the 6x86MX, the K6, or even the IDT C6 by 1997, and Intel's replacement parts, the Pentium Pro and Pentium II, were much bigger and more expensive but very little faster in real life.

In the past, the poor early performance of the 386, 486 and Pentium was temporary: later versions of these chips were excellent. But the introduction of the P6 family was less assured. Indeed, the much greater than usual media hype was a measure of the shortage of genuine technical merit in the product family - and the chaotic switches of strategy and form factor tell their own story.

The original intention was to switch the market to Socket 8 and all-internal cache (Pentium Pro). When this failed they announced Slot 1 and half-speed cache (Pentium II), and declared it a superior solution. To help this along, and to leave AMD and Cyrix out in the cold with an obsolete product range, they announced that sockets were dead, stopped development of advanced Socket 7 chipsets (notice the absurd 64Mb RAM cache limit on the TX), and started progressively adding advanced features to their Slot 1 chipsets only (notably AGP and 100MHz bus speed).

In years gone by, this would have been sufficient. This time the FUD didn't stick. Socket 7 refused to die on cue: indeed, with excellent fast new CPUs from Intel's competitors, and advanced new chipsets from VIA and ALI, it was thriving. With 100MHz Super 7 products appearing at about the same time as Intel's 100MHz Slot 1 offering, the performance advantage of the Pentium II remained small, and the cost penalty - price cuts notwithstanding - very large.

Intel's dominance of the high end of the market was not under threat — indeed, it was more secure than ever - but they badly needed a successful new mid-range CPU for the mainstream market. They kept the venerable Pentium in production long past its planned demise because they had nothing to replace it with except the very expensive Pentium II. Intel's market share dropped by over 5% in the second half of 1997, and revenue by much more. The obvious answer was to introduce a faster re-design of the Pentium Classic (if you like, an Intel K6), but they left it too late to start design work and it would have involved a massive loss of face. They had no choice but to persevere with the P6, but in Pentium II form it was still far too expensive for most buyers. This dilemma led to one of the most extraordinary product introductions of all time: the original Celeron. Here was naked cynicism at its worst - a "new" and "improved" product that was outperformed by every other chip on the market, even the old Pentium MMX it was supposed to replace. This was Intel's lowest ebb: whichever way you looked at it, the Celeron was a joke - it was large, clumsy, not very cheap, and very, very slow.

On the strength of the Celeron disaster, Intel's market share went through the floor. In 1998 their slice of the US retail market fell from over 90% to less than 50%. By early 1999, the unthinkabe had happened: they were in second place to AMD. Of the ten best-selling computer models in the US retail market, only one had an Intel CPU.

But a wounded elephant - even a badly wounded one - remains an elephant nevertheless. Turnover remained high, profits were still good (even after the massive price cuts of 97/98), but the one-time untouchable market leader was fast loosing what remained of its hold on the retail sector - which is easily the biggest growth area in the industry. They had three factors working for them, however: (1) with their massive resources they could afford to sell their lower speed parts at or below cost, (2) their still formidable consumer mind-share helped them remain strong in branded systems for the mass market where performance matters less than marketing (at Red Hill we call this "The Amstrad Factor"), and (3) their awesome manufacturing expertise. This last factor is what pulled the 386, the 486 and the Pentium out of their early high-cost, low-performance troubles, and it was manufacturing expertise that slowly started doing the same for the P6 family.

This time, though, it was a very different process. To deal with the crisis, they rushed out a revised Celeron. With a massive 19 million transistors it was a very expensive and difficult part to make, but no matter - Intel had plenty of money and they have always had the ability to work production-line miracles with silicon. And they did not just successfully manufacture it, they were soon making it at higher than expected clock speeds: it was a vastly better chip than the slug-like original, and almost the equal of recognised high-performance parts like the K6-2 and Pentium-II.

For the first time ever, Intel sold competitive CPUs for less than US$100 - not just old, slow parts, but mainstream products like the 300A and the 333. These stopped the rot. The follow-on products, released early in 1999, went still further. In order to compete with the K6-2, the Celeron was pushed so far up the performance spectrum that the Pentium II became a product without a purpose — vastly more expensive and little if any faster.

But for the first time ever they have lost their market mind-share domination. It's hard to get to the top, and it's a lot harder to stay there. But it's the hardest thing of all to stage a comeback once you've lost it.

In the longer term, Intel's massive 64-bit IA-64 project is floundering, and Merced (the first chip of this radical new family) is late and rumoured to be very sluggish. It looks as if IA-64 will not be able to take on X86 competitors (let alone the Alpha) until the second generation at least. There is a new sense of confidence among the surviving RISC chip makers, they feel they have a breathing space and are even thinking hard about developing new products, at least until IA-64 matures. AMD have now become a serious force, and should their K7 live up to its (perhaps inflated) expectations, Intel will be hard-put to maintain market leadership.

The once unquestioned market leader has made a series of bad mistakes over the last few years: ignoring the massive growth in the entry-level market, abandoning their most successful product in mid stream, then releasing the worst dog to hit the CPU market since the 80186, and now (it seems) not delivering on Merced, their great white hope. There is no doubt they are hurting. But they are learning too: they used their incredible production expertise to rush through a decent Celeron, they swallowed their pride and abandoned their exclusive commitment to the clumsy and expensive Slot 1 packaging, and (most significant of all) they are finally prepared to get down and slug it out in the sub-US$100 market - which they have to do if they are going to maintain their market share. Is it too early to say that they have turned the corner? Perhaps, but the signs are there. We expect a few more surprises, and it never pays to underestimate a wounded elephant.

As an influence on the market, it's all too easy to overrate Intel. They did not make it possible for us all to have PCs as some half-informed critics claim. If it hadn't been for Intel, we would all be running systems based on a different architecture — Motorola, MOS, Zilog, Digital or National Semiconductor perhaps. Nor have Intel made computing affordable - AMD and Cyrix made computing affordable, Intel did their best to stop it. (To be fair here we must give honorable mention also to the major hard drive and RAM makers and the host of Taiwanese component firms.) What Intel have done is speed the whole process up. Their cheap but clumsy 8088 pushed aside the more elegant 8-bit MOS and Zilog chips and took us into the 16-bit era a year or two faster than would have happened if we'd waited for the 68000 or the Z-800. And their never-ending need to sell huge numbers of fairly expensive processors has forced them to find ways to push the performance envelope out again and again.

In the poker game of market share, AMD and Cyrix drop the price, so Intel see their price-drop and raise them half-a-dozen Winstone points. Then AMD and Cyrix match the performance gain and raise them another price cut. So Intel ....

The war goes on. Intel's shareholders have done very well out of it, but you and I are the winners.