Rackable Systems slips into SGI's name

Considering all of the technical trials and economic tribulations that Silicon Graphics has gone through since the mid-1990s after a meteoric rise in technical computing in the 1980s, you might think that the SGI name was one relegated to the dustbin of the IT industry.

This seemed particularly true after Rackable Systems - an upstart niche server maker focusing on high-efficiency server designs - bought the carcass of SGI late last week for $42.5m. But no. As it turns out, it's the Rackable Systems name that is being tossed into the dustbin.

That's right. Rackable Systems is now Silicon Graphics International, or SGI for short, mimicking the same strategy that Tera Computer took when it bought the carcass of the Cray parallel supercomputer business from SGI in March 2000 for $35m. Cray's technology and name has been kept alive thanks to management by Tera, and it looks like the best that SGI's 4,500-strong global customer base can hope for is history to repeat itself this time around, with Rackable's management able to keep the SGI name alive and preserve its key technologies, such as global shared main memory and its related NUMAflex clustering. (Rackable had about 300 of its own high-end customers as well as customers who have acquired its rack-based servers through its channel suppliers).

While Rackable might be assuming the SGI name from here on out, the company's stock will still be traded on the NASDAQ exchange under the RACK symbol. And the new SGI's headquarters will be where Rackable is, in Fremont, California, not in the Mountain View stomping grounds of SGI that have been more or less taken over by Google. The new SGI resides at www.sgi.com on the Web, but is still forwarding to the www.rackable.com site, which doesn't have the old SGI's Altix 4700 and Altix ICE servers and InfiniteStorage products up yet.

The lower-case "sgi" logo of the old Silicon Graphics Inc. has been recast in the blue and green colors of the former Rackable Systems logo, which is intended to convey the merging of the companies, even though this is not really, technically speaking, a merger at all. It's a mercy acquisition.

The acquisition of SGI's assets was announced on April 1 when the original SGI went bankrupt, was approved by the bankruptcy court on May 1 for $42.5m (up from the original $25m proposed price), and completed on May 8. George Skaff, chief marketing officer at the new SGI (who had that post at Rackable), said that it had to happen fast and that the company's engineering teams are right now going over their current products and future roadmaps to sort through and merge the products. "We are adopting the best technologies from the two companies," says Skaff," and we anticipate that the majority of the product lines will survive."

While Skaff was not able to provide any details on the future of the Itanium-based Altix 4700 supers, which use NUMAflex to create a massive memory space for applications to run in, he did say that the Nehalem-based "UltraViolet" NUMAflex servers, which SGI is widely believed to have been working on for years, would eventually see the light of day. Where the overlap is high - say, between Rackable's two-socket Xeon and Opteron rack servers and SGI's two-socket Altix ICE Xeon blade servers - there is likely to be some product consolidation, according to Skaff. But again, no plans have been finalized yet. The important thing is that all current and past Rackable and SGI products will be supported as-is, and all current products continue to be for sale today, regardless of what the future roadmaps hold.

In a letter to customers, Mark Barrenechea, president and chief executive officer of the new SGI, said that Tony Carrozza will lead worldwide sales and marketing at the new company (this is the same job he had at Rackable), with David Yoffie heading up a new global services organization (he used to be in charge of Rackable's manufacturing operations), Diane Gibson running the company's internal operations (same gig at Rackable), Gio Coglitore in charge of engineering (formerly chief technology officer at Rackable), and Eng Lim Goh as chief technology officer (the role he had at the old SGI).

No matter what you call either SGI or Cray, two things remain true.

First, the U.S. government has a vested interest in keeping competing architectures alive and multiple supercomputer suppliers. But unlike in the early 1990s, when Uncle Sam propped up indigenous (and usually Unix-based) super makers through the ASCI program, these days the government doesn't have to pony up development and production seed money so much as learn to make use of technologies created mostly for the hyperscale computing of Web 2.0 companies.

That's is not to say that at the very high end of supercomputing there is not plenty of investment by Uncle Sam, such as with Linux clusters with hybrid architectures, often combining x64 and graphics processors or other kinds of accelerators (like IBM's Cell chips) into multi-petaflops machines. The table stakes are much higher than a decade ago, and very fast interconnects are going commercial. Supercomputing has, in essence, gone mainstream even as it has become less diverse.

Second, it is very important to not get stuck in niches, but this is almost impossible to avoid - particularly in supercomputing. Being a niche rack server or supercomputing server supplier are how Rackable and SGI came into being in terms of servers (SGI obviously got its start in the early 1980s as a workstation supplier, much as Sun Microsystems did and, not coincidentally, are both seeded out of Stanford University). Being stuck in a niche is what doomed SGI to failure time and again. There are almost too many examples, but here's one.

In one of the worst blunders in the history of computing, SGI sold off the "Starfire" 64-processor SMP server business it bought from Cray in February 1996 to Sun Microsystems only months later for a mere $50m. That Starfire machine and the people who came with it was the dot in the dot-com boom, and the Starfires not only made Sun billion of dollars in profits but it made Sun into a preferred commercial computing supplier. The Starfire was not so much Sun's genius as it was Cray's and then SGI's, the latter of which idiotically sold the goose that laid the golden eggs as it tried to stick with and defend its supercomputing niche.

Rackable has the impossible task of trying to go more mainstream in the server business because the server racket is a volume business these days. This is a feat that Sun itself has been trying to do with its "Galaxy" x64 server business for the past five years without much success. As Rackable has learned, any niche technology can be copied by one of the tier one server makers, who can fund their supercomputing development and custom server development from their much larger customer bases for generic servers.

This is why Cray and SGI have struggled for the past decade or more. The good news is that these to venerable companies were both given new leases on life, perhaps not the lives they once imagined for themselves, but there is something to be said for still being in the game. ®