By Matthew Dublin

The North Rhine-Westphalia Technical University (RWTH) in Aachen, Germany has chosen a bullx supercomputer to help accommodate researchers in the Engineering and Life Sciences faculties. The new system is comprised of over 28,000 processing cores and will deliver some 300 teraflops of power and three petabytes of disk storage.

The University's Center for Computing and Communication will be collaborating with Bull to optimize standard HPC applications, such as OpenFOAM, for hybrid cluster architectures that use numerous multiprocessor systems with large memory capacity together with a high-performance network. This architecture makes the most of the benefits offered by both Message Passing Interface (MPI) standards and the significant memory available through OpenMP, an application programming interface that researchers at the Aachen Center for Computing have used for quite some time.

Here's a system performance breakdown:

The overall power of the system -- at 300 teraflops (300 teraflops = 300 x 10^12 floating-point operations a second) -- is virtually the same as 10,000 of the latest desktop PCs.

Light travels at 30 centimeters a nanosecond. In the same timespan, the RWTH supercomputer will complete 300,000 operations.

The system can write up to 19 GB of data a second to the attached storage system: filling the equivalent of four DVDs.

The disk storage system has a total capacity of three petabytes, or 3,000 terabytes (3 x 10^15 bytes). It would take an MP3 player 6,000 years of continuous operation to play the equivalent amount of data.

Its processing power would rank the new computer as one of the 30 most powerful supercomputers in the world, compared with the most recently published TOP500 listing.

By Matthew Dublin

Virginia Tech University’s Kevin Shinpaugh, director of high-performance computing, has posted a video presentation on some of their Advanced Research Computing projects.

In the video, Shinpaugh discusses their Athena GPU test bed cluster that provides visualization and simulation capabilities for data intensive applications. Athena supports projects that are headed up by researchers at the Virginia Bioinformatics Institute such as a computational epidemiology application called EpiSimdemics that incorporates a Google maps-like interface with a highly scalable, parallel algorithm to simulate the spread of contagion in large, realistic social contact networks using individual-based models.

Virginia Tech - GPU Computing for Computational Sciences and Engineering from Appro International on Vimeo.

By Matthew Dublin


Argonne National Laboratory has announced that its new supercomputer “Mira” will be based on IBM’s Blue Gene/Q supercomputer technology. At 10-petaflops, the new supercomputer will be 20 times faster than Argonne’s current supercomputer called Intrepid and twice as fast the current fastest supercomputer in the world. Mira is slated to become operational in 2012 and will enable research in a range of field including life sciences.

"Computation and supercomputing are critical to solving some of our greatest scientific challenges, like advancing clean energy and understanding the Earth's climate," said Rick Stevens, associate laboratory director for computing, environment and life sciences at Argonne National Laboratory, in an IBM press announcement. "Argonne's new IBM supercomputer will help address the critical demand for complex modeling and simulation capabilities, which are essential to improving our economic prosperity and global competitiveness."

Argonne researchers are characterizing Mira as a sort of stepping stone to the world of exascale computing in that users will be required to scale their code to run on upwards of 750,000 individual processing cores-even though if the concept of exascale class supercomputers ever comes to fruition this require programmers to code for hundreds of millions of cores.

Blocks of time on Mira will be awarded to investigators through the Department of Energy’s Innovative and Novel Computational Impact on Theory and Experiment and the Advanced Scientific Computing Research Leadership Computing Challenge programs.

Before a new supercomputer is rolled out, Argonne “shakes down” the system with the help of a small community of users running production applications called the Early Science projects. The process began back in October of last year with a workshop on Mira’s architecture, configuration, and software environment, wherein project teams kicked off intensive efforts to adapt their software to take advantage of Mira’s Blue Gene/Q architecture. These projects span a range of scientific computing applications including molecular dynamics and computational chemistry experiments.

By Matthew Dublin

DRC Computer Corporation (DRC), a reconfigurable computing hardware vendor, has achieved 9.4 trillion cell updates per second running the Smith-Waterman algorithm with Affine gap model on the company latest coprocessors. DRC claims their new SW to be five times better in terms of price and performance than any other published results. The benchmark was achieved running 200 base-pair DNA reads against a 650,000,000 nucleotides database on a clustered server operating as a cloud computing environment using the SSEARCH35 tool within the FASTA genomics tool kit.

Smith-Waterman has always been regarded by the reconfigurable computing community and FPGA vendors as one of the best bioinformatics algorithms to port to an FPGA. In the last few months alone several Smith-Waterman implementations have been released, each touting more impressive benchmarks than its predecessor including SGI and Pervasive Software's Smith-Waterman implementation with benchmarks that they claimed were 43 percent faster than any current implementations at the time. Their version used a Java framework to analyze 10 million protein sequences 81.1 seconds across an 384 core SCI Altix UV 1000 for a sustained 986 billion cell updates per second.

The University of South Carolina's Heterogeneous and Reconfigurable Computing Group is one academic effort aimed at exploring reconfigurable computing for various bioinformatics tasks. So far they have adopted Convey's hybrid-FPGA solutions to explore the acceleration of phylogenetic inference methods.

By Matthew Dublin

The TeraGrid has a post describing the process their Resource Allocation Committee goes through when awarding time on the TeraGrid network for large-scale scientific computing projects. TRAC awarded a total of 1.75 billion service units — the equivalent of 200,000 years on a single processor — to roughly 1,300 applicants in 2010 for a range of disciplines, from clinical investigations of broken bones to drug design.

This year three new supercomputing systems will be added to the TeraGrid network:

Lonestar 4, a 302 teraflops Dell cluster that will go into production at the Texas Advanced Computing Center in February.

Trestles, a 100 teraflops system at the San Diego Supercomputing Center that is expected to come online in January.

Blacklight, at the Pittsburgh Supercomputing Center, an Altix UV1000 system that came online in October.

In addition, the Kraken supercomputer, housed at the National Institute for Computation Sciences, was recently upgraded to achieve a peak performance of 1,174 teraflops. In total, the three new supercomputing sites plus the upgrade will offer researchers an additional 350 millions hours of compute time.

Rommie Amaro, an assistant professor of pharmaceutical sciences at the University of California, Irvine, is running large-scale simulations for biomolecular systems on TeraGrid in effort to identify novel druggable target sites.

"We've been successful in identifying new lead compounds for several infectious diseases, including African sleeping sickness and influenza, as well as cancer," Amaro says. "TeraGrid machines enable us to perform these large-scale simulations easily and efficiently, helping us drive discovery in collaboration with experimental labs."