When Intel launched its first 32-bit processor in 1985, it might have marked a watershed moment for the personal computer industry, but users barely batted an eyelid. The 386 brought with it more than just 32-bit processing and addressing; it also brought multitasking to the desktop.
Aside from a rather high price, buyers saw only benefits in the 386 processor: it enabled multitasking for the new 32-bit applications that would follow; it ran all the old 16-bit software that had been written for the 286, and it ran that software a whole lot faster than the 286 could ever hope to manage.
Fast forward a decade or two, and some might say Intel is one bit short of a byte. Itanium, its 64-bit processor, is selling slowly however you count it. Like all chip manufacturers, Intel does not give out its own figures, but luckily for us AMD is more than happy to oblige, and estimates Intel has shipped around 16,000 of its 64-bit chips. Now you can add a few to compensate for AMD's negative spin, remove a few for the ones that Intel shipped gratis, and divide by four to get a figure that represents the total number of servers out there (few are single-processor servers) using Itanium. It's not very impressive by any measure.
The problem is, that when Intel got the 64-bit bug and began working on how it could prepare for the next big thing in processors, it turned to HP. It seemed like a good idea at the time, mind you; HP had just completed its PA-RISC processor, and was working on its EPIC architecture, to which the boys at Intel took a shine. The resulting processor would enable Intel to ditch much of the clunky x86 instruction set, but never mind: this would be an enterprise chip and enterprises would be happy to ditch the 32-bit legacy in return for all the benefits that 64-bits bring.
Now compare AMD's approach. AMD appeared on the mainstream computing scene in the early 90s with its own reverse-engineered version of the 386. After a bumpy ride through the mid-90s, caused largely by the decision to forward-engineer its version of the 486, AMD emerged with the Athlon and now the Athlon 64 -- its own 64-bit processor.
The Athlon 64, together with its brethren -- the Athlon FX51 and the Opteron -- is to 32-bit computing what the 386 was to 16-bit computing. Athlon 64 et al run 32-bit software that was written for the Athlon, Pentium and everything back to the 386 (and beyond); and early indications show they run it a whole lot faster than the Athlon can.
Like Intel, AMD doesn't like to give out figures for things like chip shipments and yield -- except, that is, when it wants to show off. And so AMD has no qualms in bragging that in just four months it has shipped the same number of 64-bit server chips as Intel has since its 2001 launch; that is, 16,000.
If you're Intel, that should be a scary number. Sure, AMD's chips are not true 64-bit in the same sense that the 386sx was not true 32-bit. But does anybody really care? I suspect not. After all, AMD's chips can take advantage of the overriding benefit of 64-bit computing -- that of being able to directly address over 4GB of memory.
With Itanium, Intel is attempting to wrench users away from 32-bit software in one uncomfortable movement. Although Itanium does run 32-bit software, it requires an emulator to do so and the speed of the resulting code is by all accounts unimpressive -- certainly slower than what can be achieved on a 32-bit processor. The strategy, which was developed over a decade ago, stands in stark contrast to AMD's easy upgrade.
AMD still has to wrestle with some uncomfortable marketing issues that inevitably turn up with any major new technology. It could start by explaining the difference between the Athlon FX-51, which is aimed at high-end gamers, and the Opteron server part: essentially there is none, except that the Athlon FX-51 is available at a healthy 2.2GHz while the Opteron currently tops out at 1.8GHz. Oh, and the server chips are a third the cost of the gaming chip (surely a first, although AMD says that when Opterons are available at the same clock speed as Athlon FX-51s, there should be price parity).
Then there is the pin-out issue. The Opteron sits in a 940-pin socket, as does the Athlon-FX 51, while the Athlon 64 uses a 748-pin layout. Confused? It gets worse. Next year, the Athlon FX-51 chip will migrate to a 939-pin package. Why? It's all apparently to do with motherboard production processes and moving the Athlon FX-51 onto relatively cheap four-layer motherboards. I'm sure it will all come out in the wash, and I think AMD can be forgiven this slight kink in the product path given the enormous implications of what it is doing here.
There's a lot hanging on AMD's 64-bit processors. For a start, the company must, if it is to return to profitability, raise the average selling price of its processors to $100. It is currently languishing at around $60 -- dragged down in reaction to Intel's relentless reductions. AMD also needed an answer to Itanium, as it has been in no position to reverse-, forward- or sideways-engineer a copy of Itanium or indeed dance to Intel's tune in any other way. But most importantly, there are a whole lot of buyers -- both consumers and businesses buying both desktops and servers -- who are rapidly outgrowing 32-bit processing. AMD might just have the answer.