Scots build green supercomputer

A supercomputer 10 times more energy efficient and up to 300 times faster than its traditional equivalents has been unveiled in Scotland.

Called Maxwell, the computer has been built at the University of Edinburgh and uses field programmable gate arrays (FPGA) in place of conventional processors.

Its Scottish developers believe Maxwell represents a new generation of compact and energy-efficient computers. Unlike ordinary general-purpose processors, FPGA chips can be programmed to perform very specific tasks. Once that programming is accomplished, FPGA chips can be much faster than performing the same tasks in software running on general-purpose chips.

The technology also has the benefits of requiring less space and running much more coolly than equivalent machines. Maxwell only takes up two computer racks at the Edinburgh Parallel Computing Center.

The main issue with the technology at the moment is that it is very hard to program. Mark Parsons, commercial director of the center, which co-developed Maxwell, acknowledged: "It's still difficult for commercial use."

But Parsons added once this problem is cracked--in the next two to three years--FPGA tech could really take off. He noted that Intel is exploring possible uses of FPGAs.

Cray, Linux Networx, SGI and others in the high-performance computing market all offer or plan to offer FPGA options in their product lines.

Maxwell has already been tested with high-demand applications from the oil, financial and medical-imaging industries. With the financial application, Parsons said, Maxwell ran at between two and 300 times faster than an equivalent system using standard processors.

Other industries that could take advantage of Maxwell include drug design, defense and seismology.

Maxwell has been developed by the FPGA High Performance Computing Alliance--led by the Edinburgh Parallel Computing Center--during the last two years. The $7 million project has been funded partly by the Scottish Enterprise, a development agency.

The computer was designed and built by Nallatech and Alpha Data. It uses FPGA technology from Xilinx.

Tim Ferguson of Silicon.com reported from London.

[CarpalDiem]

Reducing Programmability Problem

"The main issue with the technology at the moment is that it is very hard to program. Mark Parsons, commercial director of the center, which co-developed Maxwell, acknowledged: 'It's still difficult for commercial use.'

But Parsons added once this problem is cracked--in the next two to three years--FPGA tech could really take off."

Two or three years. Cracked. Right. Advances in FPGA-based reconfigurable computing utilization technologies have paced advances in AI, with steady progress over the years and lots of players declaring victory to sell their wares. In fact, given how hard it is to transform a high-level representation of an algorithm into an FPGA implementation that meets performance requirements, any effective HLL compiler will need good AI to navigate all the tradeoffs that need be made at each stage of the transformation.

Advances in FPGA designs themselves will certainly help, but getting to the point of fairly common commercial use is a gradual process of discovery and invention.