The usual winners in the high-performance computing space are massive parallel computing systems that require the area of several tennis courts to contain them, and that operate at power levels great enough to simulate the Earth's surface or nuclear blast results.

But when the newest, much-anticipated list of Top 500 Supercomputer Sites -- the benchmark of supercomputing power -- was released last week, the computer equivalent of a Honda had entered the race.

Clustered computers -- a series of smaller computers strung together to take advantage of their combined CPU power -- comprised fully 16 percent of the list, nearly doubling the number that appeared in last November's rankings.

Clusters Make the Mark

Such clusters, which even included an AMD Linux cluster at the University of Heidelberg in Germany, won top honors, with the most popular cluster of all 83 listed being IBM's Intel-based Netfinity system.

"About 80 percent of the high-performance computing area right now can be done using Linux clusters, or certainly with IBM clusters or someone else's clusters," said Bill Claybrook, research director at Aberdeen Group.

Claybrook told NewsFactor he anticipates that in the near future, most high-performance computing needs will be met by cluster computing, especially Linux-based clusters.

He acknowledged, however, that applications that require low latency and high bandwidth are not yet handled well by cluster computing.

IBM product manager Barbara Butler told NewsFactor she agrees that a switch to clustered machines is under way, even in large-scale, generalized work environments.

"In our case, we basically have moved from offering the RS-6000SP, which was initially designed as a massively parallel, highly scalable computer for those colossal workloads, to essentially building clusters out of our standard P-Series servers," Butler said.

Massive Systems Required

But it would be a mistake to rule out the massive parallel processor systems that still dominate the Top 500 list, according to Wayne Kugel, director of high-performance computing solutions at Cray.

Kugel told NewsFactor that he sees at least one critical difference between clustered systems and massive parallel machines: the ability to handle specific problems.

Such machines as NEC's Kanazawa, Japan-based Earth Simulator, with 35.86 teraflops of computing ability, and IBM's SP Power3, which has 12.3 teraflops of computing power and is used in the Accelerated Strategic Computing Initiative at Lawrence Livermore National Laboratory, both work on ongoing, specific problems -- problems that Kugel calls "intractable." (continued...)

1  |  2  |  Next Page >