INTERCONNECTS

SGI have selected SGI as its preferred technology partner for their shared-memory node, as part of the STFC funded DiRAC collaboration. This is a continuation of the national service offered by Stephen Hawking’s COSMOS Consortium, based at the University of Cambridge. This announcement enlarges and extends a long-standing relationship between SGI and the COSMOS Consortium, and will allow these interdisciplinary organisations to solve some of the most complex computing challenges of our time and find answers to astoundingly big questions.

The STFC funded DiRAC HPC collaboration has agreed that this shared memory machine is one of its highest priorities. The funding is planned as part of the recent announcement of investment by the UK Government of £158M in e-infrastructure.

“The collaboration with SGI will enable researchers using the COSMOS supercomputer across the UK to advance their ambitious programme of research in both cosmology and exoplanets,” said Professor Paul Shellard, director, Centre for Theoretical Cosmology at the University of Cambridge. “The incredible flexibility and scalability of this next-generation shared memory high performance computing solution from SGI accelerates our work exploiting large data sets and tackling complex simulations.”

The latest SGI UV system selected will include 1856 cores of Intel Xeon family processors and 14.8 terabytes of cache coherent shared memory in five racks. Intel is providing up to 32 Many Integrated Core (MIC) co-processors in this configuration, and the system will include the latest NUMAlink 6 interconnect technology, optimized for ultra-low latency data access as well fast general purpose communication.

“We have had the pleasure of working with the COSMOS Consortium since 1997,” said SGI CEO Ronald D. Verdoorn. “Now, fifteen years later and initiating their ninth generation of system upgrades, we remain committed to their ongoing success as they help unravel and understand the complexities of the origins of our universe.”

The choice of the shared memory platform has been driven by the Consortium’s need for ease of use and rapid time to solutions in this fast-moving field. One of the key benefits of a coherent shared memory system is that any program written or modified takes full immediate advantage of all cores and all memory in the system.

The COSMOS supercomputer will be a world-class Shared Memory Node within the national STFC funded DiRAC HPC Facility. The supercomputer will serve the computational needs of UK cosmologists and exoplanet astrophysicists, two of the most rapidly advancing fields in the physical sciences. On delivery it will be the largest shared-memory single-image system in Europe and it will be the first SMP system in the world enabled with Many Integrated Core technology (Intel MIC). The COSMOS system will operate flexibly as an Innovation Centre making readily available its unique capabilities to other DiRAC Facility users.

COSMOS is being actively used for the computationally challenging task of opening up new frontiers in the study of the cosmic microwave background (CMB) radiation, especially using ESA Planck satellite data. This cutting-edge SMP system makes feasible a new set of computationally challenging goals studying the statistics of the CMB. Theoretical cosmologists will also use the COSMOS supercomputer to continue pioneering the use of lattice field theory simulations to understand the physics of non-linear phenomena during the early universe. They will study cosmological models derived from fundamental theories with cosmic superstrings and extra dimensions. They will identify and calculate the observable consequences of these theories, ranging from signatures in CMB and galaxy distributions, through to the generation of primordial gravitational waves.

The COSMOS supercomputer will support key UK research in extra-solar planets. One of the main aims of this project is to be able to consolidate existing databases of astrophysically important molecules improving the accuracy and completeness of their transitions. The capability of the new COSMOS to perform such calculations will allow much more information to be extracted from spectroscopic data from astrophysical environments. This will help us answer some of the oldest questions in science such as: Are there worlds beyond our solar system? Are they numerous or rare? How many of them have the right conditions for life?

Since 1997, Stephen Hawking’s COSMOS consortium, drawn from ten UK institutions, has specified and purchased scalable shared-memory supercomputers to ensure the most rapid solutions of large-scale problems in a highly competitive, fast-advancing field. The choice of shared memory computing has always been driven by the consortium’s need for ease-of-use and rapid-time-to-solution in a fast moving and internationally competitive field. The cosmology effort specific to this technology partnership with SGI, is fully supported by the Miracle Consortium (University College London), whose studies include Exoplanets. Critical requirements to fulfill the ambitious set of science goals, including scalability of up to 16TB of globally addressable, cache coherent memory and up to 2048 physical cores are only possible through the use of SGI supercomputers.

The SGI UV high performance computing (HPC) system remains unique in the industry today in support of large single system image (SSI) systems, and with the next generation release UV will be scalable to 4096 physical cores (8192 virtual) and up to 64TB of cache coherent shared memory. In addition, SGI's new NUMA6 interconnect (50Gb/sec, 4x12.5Gb/s per direction) is optimized for global coherent shared memory, as well as global addressable memory. SGI Globally Addressable Memory (GAM) is scalable to 32k sockets.

“The ability of high performance computing systems to keep pace with the cosmological research of the COSMOS Consortium helps us all benefit from their insights on the grandest scale,” said Rajeeb Hazra, general manager of Technical Computing Group at Intel. “We are excited to work with SGI to provide an HPC solution powered by the future Intel® Xeon® processor E5 family and MIC co-processors that can intelligently adjust performance to data-intensive applications like those used by the consortium.”