NASA's new supercomp sits on a desktop
Supercomputers are bigger than the average person's holiday cottage, draw enough juice to run a small factory, and require industrial refrigeration units to prevent them cooking themselves, right?
Not necessarily. The US National Aeronautics and Space Administration (NASA) has taken delivery of a desktop-sized supercomputer which draws no more power than a hair dryer, based on Field Programmable Gate Array (FPGA) technology instead of a CPU.
FPGA chips can reconfigure themselves hundreds or thousands of times a second, making it possible for numerous applications to run simultaneously. Assuming it actually works as advertised, this setup leapfrogs over Moore's Law, yielding a desktop box with 1000 times the power of one running on a CPU.
The power boost comes from the ability of an FPGA array to maximize the use of transistors. A CPU is designed to handle many different sorts of tasks, so only a fraction of its transistors is in use at any given time. An FPGA array, by contrast, can dedicate as many of its transistors as needed for a task on the fly.
NASA's Langley Research Center in suburban Virginia is currently in possession of a HAL (what else would they call it?) "hypercomputer" from Utah-based outfit Star Bridge Systems.
Reg reader Jeffrey George reckons "hypercomputer" should be spelt "HypeOurComputer," after touring the notably fact-light Star Bridge Web site.
Additionally, reader Clark Smith seriously doubts that FPGA chips can reconfigure themselves "thousands of times a second," as a NASA press release claims. "The fastest FPGA currently operates in the 300MHz range," he writes.
No doubt we'll hear more about this faster-than-light marvel in the near future. Other HAL customers include the US Department of Defense (DoD), several telecommunications corporations and some Hollywood special-effects shops.
No word yet on how well the gizmo works, if at all; but if it's a total bust we have every confidence that Intel will promptly let us know. ®
Sponsored: Hyper-scale data management