Armed with more than 2,000 PowerPC processors, the Rochester, Minn., center is designed to give researchers and enterprises access to the supercomputing resources via a VPN (virtual private network), paying only for the compute power they use, said David Gilardi, vice president of IBM Deep Computing Capacity on Demand.
The new center joins three others—in Poughkeepsie, N.Y.; Houston; and Montpellier, France—that offer access on a pay-per-use basis to IBM systems running more than 5,200 processors from Intel Corp. and Advanced Micro Devices Inc., as well as IBMs own Power architecture.
Last year, Blue Gene topped the Top 500 list as the fastest computer in the world, with one system reaching 70.7 teraflops, or 70.7 trillion calculations per second.
To get that much performance out of the relatively small system, IBM made trade-offs in processor speed and memory. IBM, of Armonk, N.Y., has put more than $100 million in research and development into the Blue Gene project.
A Blue Gene supercomputer that IBM is building at Lawrence Livermore National Laboratory in Livermore, Calif., will reach a peak of 360 teraflops, according IBM.
Customers will have access to a peak performance of 5.7 teraflops in a compute environment designed for efficiency, scalability and low power consumption. Blue Genes footprint is less than 1 square meter.
Last fall, IBM commercialized the system—now called eServer Blue Gene—and Gilardi said the company is hoping to help build demand through the new center. IBM is giving access to Blue Gene to software makers interested in optimizing their applications for the supercomputer. Several software makers already have expressed interest in accessing the supercomputer.
Having the software makers use the system also will give IBM a better idea of the uses of Blue Gene, Gilardi said.
"We really dont know the full breadth and capabilities of a system such as this," he said. "Were trying to explore all the ways the technology can be applicable."
In addition, users interested in Blue Gene will be able to test it through the center, then decide whether they need it—either through renting the power through the center or buying a system.
Gilardi said that while most of the interest has come from the high-performance computing field, there are enterprise workloads—such as automotive design, drug discovery and fluid dynamics—that would run well on the system.