The top computer, an IBM system at the Department of Energy's Lawrence Livermore National Laboratory, is capable of 16.32 sustained petaflops, according to the Top 500 list, a global, twice a year ranking, released Monday.
This system, named Sequoia, has more than 1.57 million compute cores and relies on architecture and parallelism, and not Moore's Law, to achieve its speeds.
"We're at the point where the processors themselves aren't really getting any faster," said Michael Papka, Argonne National Laboratory deputy associate director for computing, environment and life sciences.
The Argonne lab installed a similar IBM system, which ranks third on the new Top 500 list. "Moore's Law is generally slowing down and we're doing it (getting faster speeds) by parallelism," Papka said.
US high performance computing technology dominates the world market. IBM systems claimed five of the top ten spots in the list, and 213 systems out the 500.
Hewlett-Packard is number two, with 141 systems on the list. Nearly 75 percent of the systems on this list run Intel processors, and 13 percent use AMD chips.
Despite the continuing strength of U.S. vendors globally, when China's supercomputer took the top position in June, 2010, it seemed to hit a national nerve.
President Barack Obama mentioned China's top ranked supercomputer in two separate speeches, including his State of the Union address last year.
Steven Chu, the US DOE secretary and a Nobel Prize winner in physics, warned that America's innovation leadership was at risk.
The latest Top 500 list will not change concerns about competitive threats to US technological leadership.
Just this weekend, China launched its fourth manned space mission, sending its first woman into space. The US ended its space shuttle program last year. China is also is developing its own processors to reduce its dependency on Western components.
But the US, for now, is leading the world in supercomputers.
The top system marks the first time that IBM has introduced water cooling in its supercomputers. The third place system, Mira, which is also a BlueGene system, also uses water cooling to help remove heat generated by more than 786,000 compute cores.
The Sequoia is more than double the number of compute cores of the second system on the list, Japan's K computer, which had been ranked first at 10.51 petaflops.
Along with the most first and third most powerful computer, IBM also has fourth place with a German system built for the Leibniz-Rechenzentrum computer center for Munich's universities. Two other BlueGene/Q systems, one for Italy and another for Germany, occupy seventh and eighth spots on the list.
The most striking thing about the list, said Jack Dongarra, a professor of computer science at the University of Tennessee and one of the people behind the Top 500 program, is that more than half of the machines on this list aren't deployed in research, academic settings or by government. "More than half are used by industry," he said.
"Industry gets it," said Dongarra. "These machines are important; they can provide some competitive advantage," he said.
The Europeans, in particular, are moving aggressively to build out supercomputing capability, despite all the troubles their economies are facing.
"The Europeans weren't keeping pace," said Dongarra, "and today we see resurgence (in Europe) in high performance computing."
All the European machines on the top 10 are new, said Dongarra.
IBM BlueGene/Q systems dominate the top system. The company is building systems that do not rely on accelerators and instead use its Power processors and its own interconnects, all assembled in a homogenous architecture, said David Turek, the vice president of exascale computing at IBM.
"These machines have been designed and built to solve really difficult science problems across a wide range of disciplines," said Turek.
The proof of that is in the variety of uses the IBM machines have been pegged for, which include nuclear research, earthquake prediction, life sciences, and industrial design.
Turek said architecture, rather than the brute force of processing power, is what's most important.
"The classic trick of waiting for Moore's Law to come along and help you out really doesn't exist anymore," said Turek.
The systems in the Top 500 are running processors somewhere between 2GHz and 3GHz, which is where speeds were eight to nine years ago, said Turek. "It tells you that it is parallelism and system design and architecture that carry the day here," he said.
The Mira machine at Argonne replaces a 500 teraflop system, Intrepid, which has 163,840 cores versus Mira, with 786,432 cores.
Mira is approximately 20 times faster than Intrepid and will reach 10 petaflops, or 10 quadrillion floating-point operations per second. Its chips are 16 cores versus the four cores on Intrepid.
It has 48 racks which weigh approximately two tons each. Sequoia has 96 racks. This is Argonne's first water cooled supercomputer. The transition to water was a new thing for the lab, especially for people who got into computing after the 1980s when water cooling was widely used in mainframes, said Argonne's Papka.
The water cooling has contributed to a five times more efficiency than the Intrepid machine, said Papka. Putting in the water cooling system has made the machine room "look more like a submarine," he said.
There are sensors throughout the system that monitor water pressure and any changes can trigger a shutdown, said Papka.
The system is still being tested and will begin running science later this year with it fully open to its users in 2013.
The system will be used for a wide range of scientific inquiry, and proposals for compute time will face a peer review, similar to a scientific journal. Argonne is a on a four-year upgrade cycle. It will upgrade again in the 2016-17 time period, and Papka hopes the next system is something in the range of 200 petaflops.
An exascale system - 1,000 more times powerful than a petaflop - may arrive by 2020, said Papka.