Supercomputer innovator preps for asset sale

SiCortex, an innovative maker of MIPS-based massively parallel supercomputers running Linux, has hired intellectual property managers to sell its assets. HPCwire was the first to report the story after catching wind that SiCortex had engaged Gerbsman Partners to help peddle its assets. Gerbsman published a notice on its blog …

[Robert Hill]

RIP...does this remind anyone of MassPar?

Too bad, another innovator bites the dust. In this world screaming for energy efficiency, dense server clusters, etc., a well-positioned, well-engineered product couldn't make it.

I blame the parents.

It's hard to think that the only real challenger left to x86 clusters is nVidia Tesla arrays...go CUDA!

The tombstone...for obvious reasons.

[Francis Vaughan]

Really sad if true

These guys were possibly the last company on the planet doing something interesting and innovative in HPC. Since Cray killed off work on the MTA (aka Tera), and Burton went to the evil empire, there has been a depressing world of ever bigger clusters, and not much else. Fine if your problem is a good fit, and dire otherwise. That the system was, in many ways, a successor to the Thinking Machines ideas (and some ex TMC guys contributed) gives it a special place for me.

Maybe Oracle could buy them. Oracle own pretty much all of the TMC IP now. Not that that will actually happen.

[Anonymous Coward]

Marketing Problem

They should have sold it as a search engine. Then it would have been popular.

[The_Yig]

Too awesome to succeed....

"There is no pure-play supercomputer maker that is sufficiently well run or well capitalized such that doing business with US national laboratories & universities cannot bankrupt it."

This, or something like it, was attributed to either S. Cray or Danny Hillis (if anyone remembers, please tell me...)

As Francis points out, some SiCortex priniciples worked at both Thinking Machines and Symbolics. Perhaps they should know by now that no good idea ever goes unpunished.

The Cray MTA series of machines is not gone. If you have the money, they will build you an XMT system with up to a million simultaneous hardware contexts w/ SSI memory addressing. Few people need that sort of capability, and nowhere near enough to make it commercially viable. Fortunately, the customer at Fort Mead will pay whatever it takes to get them, and perhaps someday, some sufficiently motivated Priniciple Investigator will write one into a grant proposal.

Until then, it's Piles-of-PCs everywhere!! Yayyy! (snore)