The machine began life as a, is currently the , and has only four publicly announced customers. But as IBM earlier promised would happen, the design has now become a comparatively ordinary product under the brand name eServer Blue Gene.
IBM announced on Monday that the Blue Gene will be available immediately with a starting price of $1.5 million. Monday is also the opening day of the SC2004 supercomputing show in Pittsburgh.
IBM is selling the machine in configurations ranging from one to 64 racks; each rack has 1,024 processors. A 16-rack configuration is the world record holder, able to perform 70.7 trillion calculations per second, or 70.7 teraflops, according to a convenient if imperfect speed test called Linpack.
Also on Monday, another new customer signed up for Blue Gene. The San Diego Supercomputing Center said it will install a one-rack system in December for cosmological research and other projects.
combines exotic and mainstream technology.
The processor it relies on is a member of IBM's Power family, but an unusual new member. Each slice of silicon possesses dual cores, or processing engines, and each core has dual mathematical processing units. Both cores can perform mathematical calculations, or one can be devoted to communicating with five different networks that link the processors to each other.
The vast majority of the processors run a customized, stripped-down operating system and are at the beck and call of master chips that run Linux. While that's unusual, from a programmer's perspective, Blue Gene looks just like anyand is able to run that software.
IBM has an aggressive program in high-performance technical computing, a market in which it is second only to Hewlett-Packard and is gaining share. IBM sells many products for the market and in September established a new office devoted to.
The biggest Blue Gene currently resides in IBM labs in Rochester, Minn., but will be moved by the end of the month to. Other Blue Gene customers include , the run by Astron in the Netherlands, and .
IBM launched the Blue Gene project in 2000 as part of an effort to speed calculations that predict how strings of biochemical building blocks, encoded by DNA, fold into large molecules called proteins.