My history with SGI stems from the regular swings I used to take through Silicon Valley when I was writing about microprocessors for EE Times in the 1990s. Typically, I'd stop in at SGI, where they'd tell me about the latest MIPS chip under development and talk about the upcoming workstations being built around that device. Then I'd swing by Sun Microsystems, where we'd chat about the newest UltraSparc processor. (Between the two visits, I'd grab a nutritious lunch at the Taco Bell on El Camino Real in Mountain View.)
I want to be careful not to conflate MIPS with UltraSparc, since the latter is still in production. (You can get high-end servers powered by the UltraSparc T2 and T2+, aka Niagara 2. A wide mix of other SPARC chips and servers are also available, including the UltraSPARC T1, and SPARC64 VI and VII.) For its part, MIPS was spun off from Silicon Graphics as a separate company in 1998 and is now an IP silicon house, meaning it licenses its chip cores as intellectual property.
OK, lengthy disclaimer aside, what this all means is that while the world's best chip designers were pushing the limits on instruction pipelines and doing gymnastics on silicon real-estate, the economic underpinnings of the business were changing beneath them. (This is the "being a blacksmith in 1910" meme, which is precisely what's happening in my business. As in, you're reading this content for free, but someone's gotta pay me to create it.)
So, by the mid-1990s, new microprocessors were costing multi-millions of dollars to create, and at the same time commodity x86 processors were pushing, capability-wise, into the space previously owned exclusively by high-end proprietary RISC architectures like MIPS, SPARC, HP's PA-RISC, and IBM's Power (that's Power, not PowerPC; the latter is the single-chip architecture used for a long time in Apple's boxes).
But you probably don't want a history lesson; you'd rather know why I'm waxing so nostalgic about some company you've only vaguely heard of. Well, it's because at one time graphics were tough, and really specialized. You couldn't just pop a fancy card into your desktop computer. In the days of green- and amber-screened PCs, good graphics could be found only as finely engineered (and expensive) front-end systems, usually attached to supercomputers. (The high-performance boxes did the back-end data-crunching on, for example, a finite-element analysis app, while the separate graphics box did the rendering.)
SGI was part of an ecosystem of hardware and software providers which morphed these laboratory graphics engines into standalone workstations, affordable enough for any company (though not individuals) to buy. They also helped create the electronic-animation arena, by building rendering farms to make movies like Pixar's Toy Story possible.
In retrospect, there's probably nothing SGI could have done to stanch the downward slide of the standalone graphics workstation. Still, the 1998 press conference where SGI announced its intension to start making low-end Wintel boxes is seared into my brain. Along with the long-time SGI hands at the event, you had a bunch of former Cray Research people in addition, because that supercomputer maker had been acquired by SGI in 1996. You could hear a pin drop in that room. No one, except new SGI CEO Rick Belluzzo, who was up on stage making the announcement, was into the low-end commodity strategy.
I thought to myself then and there that it was only a matter of time. And now that time has come. So long SGI, we'll miss you.
What's your take? Let me know, by leaving a comment below or e-mailing me directly at [email protected].
Like this blog? Subscribe to its RSS feed: (here)
For a microblogging experience, follow my daily observations on Twitter: (@awolfe58)
My videos on ( YouTube)
Alex Wolfe is editor-in-chief of InformationWeek.com.