Entries with tag supercomputing.

The much-discussed idea that supercomputing performance could soon reach exaflops (1018 floating point operations per second) levels will not be possible before the end of the decade, according to Horst Simon, the US Lawrence Berkeley National Laboratory’s deputy director. A combination of technical challenges are proving an obstacle, including the total power needed by such a system, increased chip power efficiency, and the cost of data movement and memory. “I also think calling the system exa-anything is a bad idea. It’s become a bad brand, associated with buying big machines for a few national labs,” he told HPCWire. “It also sets the community up for a perceived failure if we don’t get to exaflops.” And measuring the system’s performance once it is built also poses a challenge, he adds, estimating an exascale system will need five to six days to run the LINPACK benchmark. A reasonable goal toward exascale computing, Simon said, would be constructing an exascale system that could rank first in the TOP500 supercomputing-performance list by 2020. He says there are projects working in that direction, including the US Department of Energy’s FastForward. Simon says the US needs exascale computing resources to maintain a competitive advantage in manufacturing as well as for national security. (SlashDot)(HPCWire)(Scientific Computing)(“No Exaflops for You,” Horst Simon)
 

The fastest supercomputer owned by an academic institution is now online at Indiana University. uses CPUs and GPUs, and operates at maximum of 1 petaflops. Academics will use IU’s Cray-based Big Red II system in the sciences, medicine, humanities, and fine arts. Indiana firm needing help with tasks such as advanced-product modeling will also be able to work with the machine. The computer has more than 21,000 CPU and GPU processing cores and will use a new high-speed, high-bandwidth disk-storage system. IU says it is an asset that should help attract and retain faculty, particularly those whose work requires advanced data-processing power. Officials say the computing power will, for example, let researchers complete a human-genome analysis—a task that typically takes six months—in eight days. Big Red II replaces the original Big Red, a 28-teraflops computer with 4,100 processing cores that became operational in 2006. (SlashDot)(Network World)(Indiana University)
 

Scientists at the Stanford University-based Center for Turbulence Research, which is operated by both the school and NASA, set a new supercomputing record by using a million processing cores to model supersonic jet noise. They used the US Lawrence Livermore National Laboratory’s IBM Sequoia Blue Gene/Q system to solve the complex fluid-dynamics problem. Their work could not only help develop quieter aircraft engines, but also proves that million-core simulations are possible. (EurekAlert)(Stanford University)