Even as times get tough for pioneering startups building semiconductors and computing equipment, the chip industry needs to maintain its biodiversity, says Matt Reilly, a co-founder of the recently shuttered SiCortex. I wrote yesterday about the green supercomputing company selling its assets, and Reilly left an excellent comment taking me to task for calling the silicon that powers the SiCortex computer proprietary simply because it didn’t use x86 architecture. I considered that the beginning of a dialogue, and indeed, Reilly graciously agreed to talk about SiCortex’s history and his perspective on trends in the computing world.
Reilly explained that since the underlying architecture SiCortex used was a common MIPS-based one tweaked to optimize communication between the processor cores themselves and the processor cores and the memory, it wasn’t proprietary. Given that it was a specially designed chip, I begged to differ, but he is correct in his assertion that assuming anything that isn’t an x86 chip is proprietary is wrong.
When I write about corporate data centers and consumer PCs, I dub anything that isn’t an x86 chip as a specialty chip for computing. Reilly rightly pointed out that, by categorizing chips this way I basically acknowledged only Intel (and to a lesser extent, AMD and Via Technology), which helps create a monoculture for chips. Intel’s all for this monoculture, by the way — that’s why it’s building an x86 GPU and an x86 chip for cell phones. What he didn’t point out was how that argument also involves a pretty narrow definition of computing. Since I write about netbooks with ARM-based chips, using DSPs to build things like a low-power supercomputer at Lawrence Berkeley National Lab, and computers taking advantage of GPUs or architectures like IBM’s Cell for parallel computing jobs, I should really know better.
While obviously disappointed that the company he helped build and create is now on the block, Reilly tried not to dwell on the what-ifs, instead noting that SiCortex managed to build an entirely new system for faster and greener supercomputing with a relatively small team. They designed a lot of new features — from the chip architecture to the design of their box — with about 80 people, a lot of help from the open-source community and $68 million. That’s not bad, especially in light of some of the amounts raised by chip startups in the late 90s. And even a failed startup can make a mark. “You can do really interesting things with a small team today,” Reilly said. “No one was talking about green computing back in 2002 and 2003 and now it’s a big deal. And we helped set the target for that.”
SiCortex is a reminder that small players can still go up against the big guys. They may not always win, but that doesn’t mean they can’t steer the conversation.