Onsite Research
High-Performance Computer Clusters
High-performance computing infrastructure and partnerships with regional science
consortiums allow NETL scientists to determine a wide range of physical and
chemical properties – from nanoscale molecular studies to the feasibility of
commercial-scale power plants. Computer modeling provides answers where actual
experimental data would be difficult and costly to obtain. For example, computer
models simulating a proposed coal-based transport gasifier plant are helping
to address scale-up design and performance issues before the plant becomes
operational. Computational chemistry analysis allows for the screening of specific
materials for specific purposes, such as future hydrogen production.
Three computer clusters (256-CPU, 232-CPU and 128-CPU) at NETL compare to
supercomputers listed in the top 500 in the world.
- The 256-processor cluster is optimized for computational fluid dynamics
calculations with 256 Xeon 3.0-GHz processors with gigabit Ethernet interconnection.
Estimates of cluster performance based on benchmarks of similar systems indicates
that the 256-CPU cluster is rated at approximately Rmax = 961 GFlops and
Rpeak = 1567 GFlops using the LINPACK rating.
- The 232-CPU computer cluster has been optimized for computational chemistry
calculations and consists of 232 Opteron 2-Ghz processors with gigabit
Ethernet interconnection. The 232-CPU cluster should be rated at approximately
Rmax = 540 GFlops and Rpeak = 928 GFlops using the LINPACK rating.
- The 128-CPU cluster is optimized for computational fluid dynamics calculations
with 128 Dual Core Opteron processors with InfiniBand interconnection. Estimates
of cluster performance based on benchmarks of similar systems indicates that
the 128-CPU cluster is rated at approximately Rmax = 570 GFlops and Rpeak
= 922 GFlops using the LINPACK rating.
High-performance computing at NETL is further enhanced through a regional
supercomputing science consortium – SC2 – between NETL, the Pittsburgh Supercomputing
Center (PSC), and sponsors from academia and government, including Carnegie
Mellon University and the National Science Foundation.
The high-speed optical network established between NETL and the PSC terascale
computer – the “Lemieux” – is the fastest open research supercomputer in the
nation. Researchers are able to transfer gigabytes of data from simulations
completed at PSC back to NETL for post-processing and visualization in a matter
of minutes.
Additionally, 7.5 terabytes of mirrored raid level 5 dta storage
are available at NETL.
|