Parallella: supercomputing for the masses?
A Kickstarter campaign seeking to build a US$99 "supercomputer for everyone" saw its funding target of $750,000 comfortably met on Saturday, raising just shy of $900,000 in pledges. The Parallella is billed by its designers at Adapteva as an affordable, open and easy parallel computing platform based on the company's own multicore Epiphany chips.
It's perhaps more realistic to think of Parallella as a highly efficient piece of parallel computer engineering rather than a supercomputer. Though the term supercomputer is ill-defined, the word tends to conjure images of room-filling, petaflop-crunching mainframes (though the term mainframe is hardly more concrete). This ain't that.
What it is, though, is a powerful parallel computer complete with Adapteva's 16-core Epiphany-III processor which the company claims is capable of 32 GFLOPS peak performance while consuming no more than 2 watts of electrical power. The company claims that a Parallella has "10–50 times more performance" than the Raspberry Pi. If 16 cores sounds impressive then consider that, had Adapteva reached its stretch goal of $3 million, then its aim was to ship 64-core Epiphany-IV-based Parallellas come May 2013. In any case its development roadmap leads to 1000-core processors come 2014.
In the meantime, though, a $99 Epiphany-III Parallella will ship as a PC on a circuit board roughly the size of a credit card, complete with a separate Zynq-7010 dual-core A9 CPU, 1 GB RAM, two USB 2.0 ports, a microSD card reader, an HDMI connection and an Ubuntu installation (among other things). Ubuntu runs on the dual-core Zynq-7010, leaving the Epiphany-III free for whatever dark arts parallel programing wizards eventually dream up for it.
The Parallella is partly aimed at the tinkerer market, much like the Raspberry Pi and Arduino which Adapteva (a semiconductor and chip startup founded in 2008) says directly inspired it to make the computer. But Adapateva says its prime target is the developer, "the guys who really want to study parallel programing," says Adapteva CEO Andreas Olofsson in a promotional video. "We've had lots of people interested in our platform, but our price point has turned them away: thousands of dollars today is too much for somebody to do research." The Kickstarter funding, then, will be spent to tool up for mass production, to turn prototypes costing three figures into single-figure chips suitable for the mainstream.
Olofsson acknowledges that backers are doing Adapteva the favor by enabling mass production of the Parallella, and pledges that in return, the Parallella platform will be as open as possible. Software, architecture reference and documentation will be published on the web. The software will include a development kit which will include a C compiler, multicore debugger, Eclipse IDE, OpenCL SDK/compiler, and run time libraries.
In the video below Olofsson not only introduces the Parallella, but makes the case for a grassroots parallel computing movement.
Supercomputer? Possibly. Super computer? It certainly sounds like it.
Sources: Kickstarter, Adapteva
Please keep comments to less than 150 words. No abusive material or spam will be published.
Ballistic Transport of electons in computer processing applications was a hot topic of conversation in the late 1970's Terrahertz processing speeds and low power. Parallel processing is just trying to scape the bottom of an antiquated barrel. And ballistic processing won't last long either! Google already deploy quantum computing, on their Image Search, it's ideally suited to the task, as well as being 20 years ahead of expectations. We thought quantum would not arrive till after 2030!