FROM EDN EUROPE: Standard FPGA hardware will power Teraflop computer

A consortium based in Edinburgh, Scotland is to build and operate the largest computer so far planned, based on reconfigurable computing techniques and on FPGA hardware. The machine will be a 64-node configuration and will be in the "supercomputer" class, aiming for operating speeds in excess of one Teraflop. It will be built with COTS (commercial-off-the-shelf) hardware. The project is planned to take two years and will be based at the Edinburgh Parallel Computing Centre (EPCC) at Edinburgh University, which already hosts Europe's largest "conventional" supercomputer. Partners in the alliance include FPGA IP specialists Algotronix; reconfigurable systems suppliers Alpha Data and Nallatech; Xilinx; and the Institute for System Level Integration, a joint research initiative between several universities in Scotland.

A computing system will achieve the fastest execution of any algorithm when the algorithm is directly implemented in hardware. An FPGA-based computing system offers a way to approach this ideal, using reconfigurable hardware, while retaining a general-purpose computer's ability to tackle any computation problem. The principle has been known for a long time, and commercial companies (Nallatech and Alpha Data among them) have used it to build successful product ranges. This is believed to be the first time that the technique will have been applied on such a large scale.

One impediment has been the difficulty of achieving the same flexibility as a conventional computer in handling any arbitrary problem. For a conventional computer, code is written and compiled. For a reconfigurable hardware machine, the user has to partition the problem to determine which parts are best suited to running in hardware and software; then generate the configuration code for the hardware components, and compile the software code. The alliance will aim to demonstrate the progress that has been made in that process by selecting and porting three existing supercomputer applications to the new machine. From this work, the partners will create tool kits and other IP and make them available for license, and it is planned that this activity will become the centre of a new "cluster" of FPGA-based computing activity. Once running, researchers with computation-intensive academic projects will be able to reserve time on the computer.

EPCC, www.epcc.ed.ac.uk;

Algotronix, www.algotronix.com;

Alpha Data, www.alphadata.co.uk;

ISLI; www.sli-institute.ac.uk;

Nallatech, www.nallatech.com;

Xilinx, www.xilinx.com