In preliminary performance tests carried out on 2,112 of the system's 2,200 processors, the so-called "Big Mac" cluster achieved 8.1 teraflops, or trillions of operations per second, according to figures published on Wednesday. The system is still being tuned, and final results won't be announced until next month, but the performance figure would place the Big Mac at No. 4 on the list of the world's fastest 500 supercomputers.
The figures are remarkable partly because Macintosh hardware has long been absent from the top 500 list, but also because of the Big Mac's cost. In a world where the top machines traditionally cost $100 million to $250 million, and take several years to build, the Mac-based system cost just over $5 million, and was put together in about a month.
Virginia Tech said the final performance figure could be much higher. The 8.1-teraflop figure is only 48 percent of the system's theoretical peak of 16.8 teraflops, and it may be possible to squeeze more efficiency out of the cluster.
Among the three machines ranked above the Virginia Tech system, Japan's top-ranked Earth Simulator runs at 87 percent of its theoretical peak, and the other two run at 67 percent and 74 percent. The figures were posted in a report by Jack Dongarra, a University of Tennessee computer scientist who maintains the top 500 list. Earlier tests using just a few of the Big Mac's processors reached roughly 80 percent of the theoretical peak.
Virginia Tech plans to use the cluster to perform research on nanoscale electronics, chemistry, aerodynamics, molecular statics, computational acoustics and molecular modeling, among other tasks.
The servers in the cluster are connected through 24 high-speed Infiniband switches from Mellanox Technologies. Infiniband, which was developed by a consortium of server and storage companies, provides greater bandwidth than other technologies on the market, such as Miranet, and can often cost less. The cluster also uses a cooling system from Liebert, a division of Emerson Network Power, as well as Gigabit Ethernet switches from Cisco Systems.
Clustering, which involves linking hundreds or thousands of computers to tackle massive projects, has opened the supercomputing market up to companies other than those such as IBM and Cray that have long made supercomputers. Dell has emerged as one of the leaders in selling clusters to research institutions such as Cornell University. Utah's Linux Networx, meanwhile, has won contracts to install systems at Los Alamos National Laboratory and other national research laboratories.
Japan's Earth Simulator, with 5,120 custom processors, was measured at 35.8 teraflops last year, and is estimated to have cost up to $250 million. ASCI Q, a Hewlett-Packard machine running on 8,192 Alpha chips, is ranked No. 2 at 13.8 teraflops.
The third-ranked system on the official list is, like the Big Mac, a cluster: it was built by Linux Networx for Lawrence Livermore National Laboratory from 2,304 2.4GHz Xeon chips, and runs at 7.6 teraflops. Another HP-built machine powered by Intel's Itanium 2 processors has not yet officially entered the list, but it would rank above the Big Mac, at 8.6 teraflops, according to Dongarra's figures.
Matthew Broersma of ZDNet UK reported from London. CNET News.com's Ina Fried and Michael Kanellos contributed to this report.
