Farewell then, Alpha – Hello, Compaq the Box Shifter
End of an Era
Analysis Digital Equipment Corporation may have officially ceased to be in 1997, when it was taken over by Compaq, but Ken Olsen's great dream really died today. Compaq's decision to divest itself of the world's fastest chip, and to standardise its enterprise servers on Intel's Itanium marks the end of forty years in which its destiny was in its own hands.
The news means that the number of enterprise RISC processors that will be in active development at the end of the decade narrows down to just two: Sun's UltraSPARC family and IBM's POWER series. Both MIPS Inc and Hewlett Packard have long and graceful retirement plans for their processors which involve several more revisions. But both HP and MIPS' largest server customer Silicon Graphics have said their RISC systems will eventually be superseded by IA-64.
Any brief hopes that Intel was acquiring the Alpha team and technologies to provide a fall-back in case the Itanic sank were dashed by today's announcement.
The announcement involves Compaq honouring its commitment to produce a next generation, SMT (multi-threaded) Alpha, after which time the engineers will be transferred to Intel projects.
In other words, they'll be shovelling coal in the Itanic boiler room.
Compaq's OpenVMS, Tru64 Unix and its Tandem NSC Unix will be ported to IA-64 by 2004. That's a colossal task, and to meet this date we wouldn't be surprised to learn that porting work has already been underway for some time. Compaq acquired Tandem a year before it swallowed Digital, in 1996, and its Himalaya servers are currently being ported from MIPS to Alpha. CEO Mike Capellas in a memo dated 12 June, and leaked to semiconductor site The Inquirer, talks of double-digit growth targets for Himalaya. Given that Compaq will be supporting Himalaya on three platforms by 2004 - the original MIPS, the interim Alpha and the new Itanic flavours, that's very optimistic indeed. Who'll want to invest in an Alpha-based Himalaya that won't be upgradeable to Itanic?
More significant than the Alpha divestment is Compaq's reorientation around being a services and solutions business, trailed in the Capellas memo. Capellas picked out Microsoft .NET as a leading candidate for future Compaq business.
We've seen this trend - dropping the home-grown technology in favour of becoming a Microsoft VAR - from many former enterprise giants over the past decade. It only remains to be seen exactly what kind of has-been Compaq will eventually become. As the example of Data General showed, it's a precarious existence. DG was swallowed by EMC some eighteen months ago.
Not having your own chip doesn't necessary mean becoming a Dell style box-shifter. But it is harder to justify continuing the investment in high end technology such as clustering in the commodity Intel model. Compaq doesn't turn into a Bull - which simply rebadges IBM RISC kit - overnight. Perhaps Unisys is a good example: it relies on Intel for chips, and Microsoft for the OS, but has used its historical mainframe partitioning expertise to produce a very fast enterprise class Windows x86 that's scooped the top spot in TPC shoot-outs. Other enterprise survivors don't provide much cheer. Fujitsu-owned ICL has hung on, just about, through its historical lock on certain vertical markets. But we can't remember the last time our new technology from ICL set our pulses racing.
DECquiem for a dream
Given that the Alpha processor's fortunes were in the hands of a PC company, and the PC business is facing a price-cutting bloodbath, the economics for continued investment in the chip were looking brutal. Compaq had fewer OEM customers for Alpha (although Cray uses the chip) than any of its RISC rivals.
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. ®