PORTLAND, Ore. — The Department of Energy's Lawrence Livermore National Laboratories (LLNL) has partnered with Cray and Intel to create a novel cluster supercomputer appropriately named Catalyst, because it will serve as a model for future procurements of high-performance computers (HPCs) by the DoE. LLNL, Cray, and Intel all contributed to the design and financing of the cluster computer and will share its resources. The project will be managed by LLNL's High Performance Computing Innovation Center (HPCIC), which specializes in working with industrial partners.
Catalyst supercomputer at Lawrence Livermore National Laboratories to serve as model for future DoE high-performance computing procurements.
"By pooling resources, the partners have jointly created a unique computational resource," Fred Streitz, director of HPCIC told EE Times:
Access to this resource among the three partners is apportioned according to the share of total cost each has contributed. As no partner requires the full resource all of the time, this arrangement allows each of the partners to explore data analysis techniques and algorithms at a scale that would be impossible individually.
The 150-teraflop supercomputer uses Cray's CS300 chassis and Intel's latest Xeon E5 processors in a unique cluster configuration consisting of two scalable units with a total 324 nodes, each with dual Xeon E5 processors with 128 gigabytes of dynamic random access memory (DRAM) and 800 gigabytes of non-volatile solid-state memory. Interconnection networking is provided by Intel Quad Data Rate optical fabrics. A dozen Intel Lustre router nodes round out the unique memory configuration by each providing an additional 128 gigabytes of DRAM and 3.2 terabytes of non-volatile solid-state memory.
Streitz told us:
The partners will take advantage of Catalyst's unique memory configuration to explore different approaches and algorithms for addressing a wide variety of data analytics and modeling issues both inside and outside of the DoE mission space. In addition, by partnering with LLNL's HPC Innovation Center, the companies can also investigate the utility of machines like Catalyst for developing solutions to industrial applications.
These essential collaborations accelerate innovation by bringing together hardware and software engineers, computational scientists, applied mathematicians, and computer scientists to develop next-generation solutions for pressing problems in such diverse areas as bioinformatics, energy distribution, and national security.
Catalyst will also be used to discover the architectural and technological features necessary to provide the analytical methods that meet the DoE's future needs, and which will subsequently be used as a model system for future HPC procurements over the coming decade.
— R. Colin Johnson, Advanced Technology Editor, EE Times