Slashdot Mirror
SGI Arises From the Ashes
eldavojohn writes "Six months ago, Slashdot reported on SGI's filing of Chapter Eleven Bankruptcy. I wondered why Slashdot kept the Silicon Graphics category with them now defunct. But Chapter Eleven means a reorganization — not liquidation. And, surprisingly, SGI has dusted itself off and stood back up. What did they dust off? About $150 million worth of spending a year. Will this reorganization put them back as a player in the graphics game? Maybe but as the article notes, they have some stiff competition that offer comparable services for less money. Is this a phoenix story or the final death throes of the company?" To be honest, no one here suspected a thing. We just keep the old topics around so it's still possible to find old stories related to them. Sometimes (like now!) they even still come in handy.
FYI, they EOP'd the Prism deskside in June 06 (it was IA64-Linux); their 'cheap' system was $7K (ridiculous.) The larger Prism systems were interesting- they were basicially a large Altix with graphics pipes strapped on; but was a solution in search of a problem. How many people need to visualize a half-terabyte of data from RAM? The demo they liked to show at trade shows was to visualize every part of a Boeing 777 (down to each rivet) in real time. It didn't wow you because you can't see every rivet (even at 10Kx10K), and it wasn't textured. If you need to explain with more than 3 words why your demo is awesome, then your demo isn't awesome.
The Altix is better in just about every category than the SuperDromes (price, performance, units shipped, IO, scalability, etc.). The nice thing about the Altix versus the Tera/Cray system is that code written by Joe Researcher on his 2P Linux desktop machine will run on 2048P Altix w/ just a recompile. While IBM's Blue Gene & Red Storm are 'linux-based', developing for the platform is nontrivial. Of course, if you're dropping $50M, you could probably swing a few dollars for some experts to optimize for that platform. They also got screwed by the Intel's Montecito delay.
SGI isn't selling Opteron clusters (They have a 'special' relationship with Intel.) They are selling Xeon clusters (commodity currently, coming out with more special sauce platforms). It's probably too late. If they came out with clusters in '99 - '01 when there were a significant SGI user-base that would pay a premium for their tools and environment, they could have captured a good share of that market.
Going Chapter-11 freed up cash. They aren't going to compete in graphics, but they have enough interesting hardware and low expenses to carve out a niche market. The ex-creditors own much of the new stock.