The PC maker on Tuesday said it is working with Pennsylvania State University to build a new high-performance computing , a group of computers linked together to perform heavy-duty computing tasks.
Penn State will use the new cluster, which currently has 80 nodes and will grow to 256, to carry out research in biology, chemistry, physics and meteorology. Each of the 80 nodes is a Dell PowerEdge server with dual Xeon processors from Intel and RedHat's Linux operating system.
The university has become the latest customer to select Dell's cluster configurations as a way to gain high-performance computing for less money. Dell also recently unveiled a 2,008-node high-performance computing cluster at the State University of New York in Buffalo.
The Round Rock, Texas, computer manufacturer is using these clusters to get foothold in the high-performance computing market, once dominated by supercomputers. The Buffalo computer cost about $4 million to build, while some supercomputers can cost in the hundreds of millions.
Clusters allow Dell to build the basic computer out of regular servers using standard Intel processors and off-the-shelf software and then focus its research and development efforts on tuning the nodes to work together efficiently. The result is that companies or institutions such as Penn State are able to gain supercomputer-like performance on the cheap, while Dell is able to sell more servers.
High-performance computing clusters can offer competitive performance for as little 10 percent of the price of a proprietary system that uses custom-built processors and custom software, said Russ Holt, vice president and general manager of Dell's Enterprise Systems Group.
Researchers, though, note that clusters don't always replace supercomputers. Clusters are handy for complex yet repetitive tasks such as analyzing genes or proteins, while supercomputers remains superior for weather prediction and scientific simulation, problems in which each calculation can have a ripple effect on others.
Penn State's Vijay Agarwala, director of graduate education and research services, said that Dell's cluster pitch, especially when it comes to large clusters, isn't just hype.
"Instead of professors deploying small, eight-node clusters in their offices...it's far more productive to build larger machines," Agarwala said. "Larger machines, when properly run, lower the cost of ownership."
All told, Penn State's three clusters can deliver a total of nearly 1 trillion calculations per second, or 1 teraflop, of maximum performance, according to Dell. But Dell has built other clusters, including one for the Buffalo campus, that deliver 5.6 teraflops, rivaling some of the in the world.
Dell is pursuing customers in areas such as oil and gas, the automotive industry and entertainment, where companies use clusters to render graphics for movies and television. It began its high-performance computing program in earnest earlier this year, with a partnership with supercomputer manufacturer Cray and a new set of cluster products.