| Sumario: | The IBM Blue Gene project, started in 1999, called for the construction of a "massively parallel" computer with over 36,000 processing chips and is now being manufactured and tested with some changes. IBM'S intention was always to apply what it learned in building Blue Gene to its commercial supercomputers. So the project is now pursuing two different architectures. The first, called Blue Gene/C, is the original clean-sheet design. The second, called Blue Gene/L, is a new general-purpose, scaled-down architecture that draws on existing commercial technologies to produce a machine which, all being well, will be running by late 9.oo4 at around a third of a teraflop (a trillion calculations per second). Blue Gene/L, which is being built in conjunction with the Lawrence Livermore National Laboratory in California, will serve as a test-bed for a number of Blue Gene technologies. In particular, it will allow IBM to test the "message-passing'' scheme used to send information from one processor to another. Depending on how well it works, IBM will then draw upon the C and L architectures to build a much faster machine, known as Blue Gene/P. Blue Gene/L is similar enough to existing Linux clusters (which are an increasingly popular form of supercomputer) to be easy to program, but powerful enough to attack entirely new kinds of problem.The Blue Gene project has become more ambitious than originally planned. But thanks to Linux, it has also become more practical, and more likely to produce commercial spin-offs.
|
|---|