主题：【文摘】一篇观点比较独特的分析文章。 -- Highway

AMD design will kill competition

THE X86 ARCHITECTURE is soon to be pretty much the only game in town. It was bound to happen. In Intel's perfect world, everybody would be using its processor architecture. And that corporate dream is being realised right now, though not in the way Intel might have hoped. It's going to have to share the credit with arch-rival AMD.

It all comes down to 64bit computing. Intel tried to stave off the market's hunger for a cheap 64bit processor, at least until it was ready to produce cut-down versions of the Itanium, but the path was set by IBM and Intel many years ago. Intel failed to pay enough attention to its own history; the lessons were there in the move from 16 to 32bit.

History Repeating Itself

In 1981 IBM announced the 5150 PC. It was the machine set to revolutionise business computing but it had a major design flaw. For some inexplicable reason IBM chose to use the 8088 processor in that first PC. It was a choice that bordered on the bizarre. The 8088 was the bottom of the line of the 8086 series of processors from Intel and most experts agreed it was one of the worst processor designs on the market. Its memory management has been described as "brain damaged" and register allocation for data was like a game of Russian roulette.

These days Intel has a reputation as a company with brains and savvy, back in the early eighties it was a different story, especially when it came to the 8086 series. To give you an idea of just how bad things were, you need only look at the 80286. You could switch modes on it to overcome some of the brain damage of the 8086. Unfortunately, once you switched modes, there was no instruction to let you switch back again. That meant all of your old programs would have stopped working if you used the new mode. Luckily for Intel, this example of its architectural skill was supplemented by an example of its implementation skill: there was a bug in 80286 that meant you could switch the mode back in software. That bug was so important and used so often by programmers that Intel had to preserve it in future processors for compatibility.

Now it might seem that 8086 series had nothing going for it at all. Here was a 16bit processor that was little more than a kludged up 8bit processor and so bad that almost nobody loved it. But that turned out to be an advantage. Where programmers on competing processors were happy to use assembly language, getting anywhere with an 8086 meant a decent compiler was essential. Compiler technology came on in leaps and bounds.

Today's World

These days the x86 architecture has taken over. Thanks, in part at least, to good compilers being available, programmers didn't have to worry too much about the dreadful architecture underneath the bonnet of a PC. The x86 has become the most common desktop processor in the world. It has banished many other, far superior, processor designs through sheer clout of numbers. And the market is moving relentlessly towards the PC as its only architecture.

Where a few years ago buying a workstation meant hunting down the likes of Sun, Silicon Graphics, HP or DEC, these days it's rare to find a buyer who doesn't investigate whether the job can be done on a PC. It's simple economics: why pay ten grand when a PC can probably do the job for two or three?

The trouble is, for many applications in the workstation world, you need 64bit. So the PC was out of the question until AMD introduced AMD64. It's a technology that was originally called x86-64. Where the Intel 80386 took the old 16bit 8086 architecture and made it 32bit, AMD64 pushes it the next step onwards.

Where Intel had hoped that its Itanium technology would finally rid it of the technically rather embarrassing x86, AMD has come along and given the nasty old throwback another decade of life.

Big Tin

The biggest obstacle to the PC taking over completely has been the top end of the market, the big tin. Until recently you would have had to start buying specialist equipment if you wanted a supercomputer. Then came PC clusters. They weren't ideal; trying to get a bunch of 32bit processors to do what is at least a 64bit job is obviously less than optimal. But clustering works and it's cheap.

When AMD launched the Opteron, it made cheap clusters of 64bit machines a real possibility. The writing is now on the wall for pretty much every other 64bit processor, with the possible exception of IBM's Power series.

That Sun has picked up the Opteron shows that the firm has realised where the market is going. SPARC might well be a good processor but, in the long run, it's had it. Getting a firm to cough up for a five thousand dollar CPU is going to be difficult when there is an Opteron that's just as fast available for a fifth of that price. Sun is playing the game reluctantly, it doesn't want to lose SPARC, but the processor is effectively already dead.

The same goes for Itanium. It doesn't really matter how good the Itanium is, big tin has now been commoditised by AMD. HP and Intel killed the Alpha processor but there are many who would argue that Alpha never came close to its potential anyway. MIPS steadily lost mindshare at the same time as Silicon Graphics and PA-RISC was effectively killed by HP.

There are those who will argue that big tin will always need specialist processors, that there are some tasks that clustered PCs simply aren't cut out for. That might be true but it's unlikely. Ten years ago nobody would have believed that Cray would be selling clustered PCs but, now that the Opteron is here, that's exactly what Cray is doing. It's cheaper for the customer, lower risk and far easier to repair if something goes wrong. It might not be a perfect supercomputer, exactly matched to the job at hand, but it's far more flexible.

Small Fry

The only thing that stands in the way of x86 architecture becoming truly pervasive is the needs of the mobile market. Believe it or not, despite all of the millions of PCs sold, x86 is not the world's most popular processor architecture. That accolade falls to a design first put together by a few guys at a small firm in Cambridge, England - the ARM processor architecture.

ARM processors appear in all kinds of things but its most notable success must be in the mobile phone. It also appears in printers, PDAs and just about anything else that needs processing power. Its big advantages over the PC architecture are that it needs very little power and very little support circuitry. The x86 simply cannot do what the ARM manages.

Conclusion

So what we have is a situation where x86 desktop processors are likely to take over the whole desktop and above markets. A 64bit x86 is a long way from being the perfect processor; it's a kludged 64bit architecture built on a kludged 32bit architecture built on a kludged 16bit architecture that was designed to be compatible with the 8bit 8080. Hardly the best choice.

But economy of scale means that it will win. Intel is having to prepare a 64bit x86 processor. It had no choice; big customers like Dell will have been breathing down its neck for years over the AMD64. It's certainly no coincidence that Dell was dropping big hints about an Intel 64bit x86 processor a short while ago; Dell is feeling the pinch of customers wanting 64bit PC compatible machines and not being able to supply those needs.

What's more, Intel is stuck with using AMD's design because so much has already been done for it. Microsoft has an AMD64 version of Windows well into beta and it's not likely to take any lip from Intel about slightly different versions - Microsoft needs this version of Windows to be thoroughly beta tested by thousands of users before it is released and they can only do that on AMD processors so Intel will have to toe the line. Then there is all of the AMD64 work done on Linux, Intel wouldn't be too popular if it caused a whole lot of rework. Essentially, Intel has to produce something that is fully AMD64 compatible or it will be making the biggest marketing blunder in its history.

The next decade is likely to see the AMD64 based PC become almost omnipresent. On everything except mobile devices, it fits every need. The grand shake out of other processors has already begun but now the pace will accelerate. In many ways it will be sad to see so many far nicer architectures disappear. At least there will be less fundamental change in the industry a decade from now, the x86-128 will be much less of an upset.