LONG: Remember, it's "the attack of the killer micros."
My comments follow. A lot of them.
theregister.co.uk
"But many will be mourning the fate of what remains the world's fastest chip. And what in likelihood will still be the world's fastest chip in a year's time, too. The Alpha's clean design is much admired by rivals, its relegation to the margins shows as great a loss of nerve, and confidence in one's own technology, as we can remember. ® "
The real issue is that CPU design has shifted to the chip companies. A company without its own massive chip R&D and production facilities (for learning curve reasons) does not have the resources to pioneer new ISAs (instruction set architectures).
Sure, they can *use* the CPUs from the chip companies in new ways...parallelism, fault-tolerance, fancy sheet metal enclosures, added software, etc., but they cannot sit down and design a new architecture and build it in large enough quantities to get the advantages that a chip company has.
(Might this change in the future, given wider use of foundries and better CAD tools and so on? Maybe. But there are few signs of this. Trends are currently in the other direction.)
And even building new box architectures using the chip-level ISAs is increasingly influenced by features put into the chips. SMP and cache, obviously. Gone are the days when a computer company simply "rolled its own" on an architecture built around a CPU. (This ended with the IBM PC, probably.)
Dell seems to be doing best in riding this wave: few fancy and proprietary (and money-burning) innovations, just plain old "use what Intel gives us." Companies which try to add their own subtle twists are risking going off the "industry-standard" track.
These trends were basically forecast many years ago. The paper "Attack of the Killer Micros" outlined the trends back in the 80s. The author concluded that the economies of scale of VLSI production ensured the ultimate dominance of microprocessors (a misnomer, as we all know, as there is nothing "micro" about them anymore, except in feature size).
DEC was caught up in these trends as PCs and RISC-based workstations pushed up from below. Their "MicroVAX" was never a mainstream chip, so it lacked the economies of scale and the software support. And the relenteless proliferation of variants and improvements. Other computer companies were also driven out of the business...
NCR started only selling PCs. CDC tried to get into exotic technoloiges to end-run Intel and the chip companis, but failed, and is now kaput. Burroughs is gone (isn't it?). A couple of the other "Seven Dwarfs" merged and are limping along as PC repackagers and VARs. Data General is gone.
The new crop of "let's use VLSI to build new architectures" companies like Convex, Alliant, Encore, Ridge, etc. are all either gone or got absorbed into larger companies...and then "Borg-assimilated." (Some are still selling specialized architectures, as with IBM's fault-tolerant machines, etc.)
Even Cray Research got bought and "parted-out." (Sun took some of the assets, sold some to SGI, etc. SGI has now switched to Intel, of course. And the Cray machine using Alphas is on a doomed course.)
All gone. The Seven Dwarfs (or Dwarves) are gone or absorbed. "The power of 2" was not enough. "The power of 7" won't be, either.
IBM is still going strong, and has extensive chip production capabilities. Will it be enough? We'll see. And even they are obviously in the "industry-standard" camp...a point too obvious to dwell on. Covering their bets...
The trend is pretty clear: the industry-standard architecture wins out *unless* there are compelling reasons for alternatives. And these compelling reasons for alternatives rarely withstand serious scrutiny...and another couple of years of the learning curve at Intel and other chip companies. Any minor performance edges are usually lost when the improvements in the mainstream processors come out.
"No one ever got fired for using Intel, except at Apple" may be the new motto of our age.
An excellent essay by a former top Motorola designer Nick Tredennick, later of Altera, appeared about 10 years ago in "Microprocessor Report." Paul Engel and I both read it (at "Computer Literacy," of course!). Nick outlined the facts of large-scale production of VLSI microprocessors, noting that Intel would produce more 386s before _lunch_ on a typical day than companies like MIPS would produce in an entire year.
(The "big" MIPS chips of the day, that is. Today, the MIPS chip is primarily a microcontroller, for video games and the like. By not expanding it to current state of the art capabilities and by using standard shrinks, it can now be made in huge quantities. So the above comparison isn't as apt--it may even be that more MIPS microcontrollers can be cranked out in a day than Intel builds PIIIs in a day: numbers of Nintendo and Playstation boxes versus numbers of high-end PCs, for example. But the point would apply to the UltraSPARC III, or whatever, or the Alpha EV, etc. More PIIIs and P4s produced before lunch than these other chips see produced in a year.)
In closing, any company thinking about basing a product around a custom CPU (custom at the chip level) has to reject such a plan. The trends are utterly obvious. Maybe someday someone will break the trend.
For the foreseeable future, the future of computer design lies in the largest chip companies, mainly Intel.
--Tim May |
