Supercomputers are generally large collections of processors that work simultaneously on solving complex problems, such as disease cures, global warming, natural-phenomenon analyses, and so on. Naturally, the higher the combined computing power, the faster the solution is found. IBM's Power7 microprocessor, which was officially detailed back in August, is set to be used in the construction of a new supercomputer, codenamed Blue Waters, which will be based at the Urbana-Champaign campus of the University of Illinois.
The high-performance computing platform will make use of the octo-core Power7 processors in order to create a working supercomputer capable of an initial peak performance of 10 petaflops. Such a high amount of floating point operations per second won't be the constant rate of computing, however, with the sustained real-world performance being estimated at around one petaflop.
The Power7 processor is based on the chip used in the construction of the Roadrunner supercomputer (which was ranked the number-one most powerful in the world multiple times by the Top500 list). The conglomerate is set to become operational next year and will have a theoretical peak computing capability of 16 petaflops. This will supposedly be achievable by connecting up to 16,384 Power7 nodes, but IBM doesn't expect the initial performance to surpass the aforementioned 10-petaflop speeds. Heat will be kept in check through a water cooling mechanism devised by the IBM engineers.
“We actually went a bit further environmentally. We took a lot of the infrastructure that's typically inside of the computer room for cooling and powering and moved the equivalent of that infrastructure right into that same cabinet with the server, storage, and interconnect hardware,” Ed Seminaro, an IBM fellow who is involved with the University of Illinois project, said. “The whole rack is water-cooled. We actually water-cool the processor directly to pull the heat out. We take it right to water, which is very power efficient.”
The supercomputer will not only provide revolutionary computing speeds, but will also boast a good node intercommunication, which will make it appealing to organizations like DARPA and the NSF.
“The transfer of data between any of those two nodes in the system is at the full rate of 192GB per second--peak,” Seminaro added. “So, you can get data from anyplace to anyplace at that kind of speed with latency on the order of less than one microsecond.”
IBM also intends to employ the Power7 processor in commercial server products, which will be marketed starting with the first half of 2010.