IBM’s ascension to the top of the world’s supercomputing ranks marks the emergence of a platform that would make its way into a wide variety of customer environments, executives said Wednesday.
The results of a benchmark test conducted
a few weeks ago confirm IBM’s Blue Gene/L knocked Japan’s Earth Simulator from its world record pedestal as the most powerful high-performance system as measured in teraflops. The Earth Simular, which was designed to study climate control, has held the record for the last two years, but IBM said Blue Gene/L’s sustained performance of 36.01 trillions of calculations per second eclipses it.
Blue Gene/L uses a power-efficient design that would scale to performance, starting with two chips on small card. These nodes are added to boards, which are in turn added to racks in the supercomputer. An early Blue Gene/L implementation, for example, will include 64 racks with 16,000 processors. The Earth Simulator, in contrast, uses fewer processors, but they are faster and more specialized.
While the benchmark tests earns IBM — and by extension the United States — supercomputing bragging rights, IBM vice-president of deep computing Dave Turek said the company is more interested in serving customers like Lawrence Livermore National Library, which is expected to install Blue Gene/L early next year.
The Earth Simulator was a government-funded project focussing on a narrow set of problems, as opposed to IBM’s more commercial aspirations, Turek said.
“”We absolutely don’t engage in science fair experiments for the purpose of beating benchmarks,”” he said. “”What we try to do is build systems that resonate in the broad set of the marketplace.””
IBM’s technical industry may position the U.S. as a high-performance computing (HPC) leader, but in Canada there is a growing concern that local industry needs more development. Last week, for example, the Canadian Advanced Technology Alliance (CATA) released the preliminary results of a HPC survey conducted on behalf of the Canadian High Performance Computing Collaboratory (C3), which indicated that many in the high-tech community have a hard time defining what HPC even is.
Kevin Wennekes, CATA’s director of policy research and advocacy, said C3 will soon launch a plan that will act on some of the concerns raised in the survey in order to foster a community that could build supercomputers that compete with Blue Gene/L.
“”Canada is very recognized for the work they do there, but of course part of problem or challenge is that no one in Canada seems to recognize that,”” he said. “”There isn’t a lot of communication, especially at the private sector level, of what’s going on in HPC among themselves. Synergies aren’t being shared, the partnerships that could help each other aren’t being formed.””
While most Canadian supercomputing projects have been affiliated with the research facilities at various universities, Turek said Blue Gene/L has already been pre-sold to a number of customers, with a broad spectrum of applications. Blue Gene/Lis intended to become a general-purpose supercomputer.
“”Although we’ve talked about it today in a very large sense — and certainly it the context of Lawrence Livermore, a very large sense — one of the beauties of Blue Gene is that scales down as well as up,”” he said. “”One can buy a very small Blue Gene system as well.””
If Canada wants a more vibrant supercomputing industry to go after the same customer base, Wennekes said the industry must find more people to train for HPC-specific skill sets.
“”Any education campaign will certainly be geared towards the youth of today to get them interested in HPC and make them realize the potential here in Canada,”” he said. “”We have to make sure we’re not losing talent to anywhere else in the world and cultivating our own talent here.””
In the most recent Top 500 list of the world’s most fastest supercomputers, Simon Fraser’s WestGrid was the highest-ranking Canadian machine, reaching No. 34.