Overview of Deepthought2


Deepthought2, which is housed at the Cyberinfrastructure Center, is the University of Maryland's flagship high-performance computing (HPC) resource. The cluster is factory rated at 300-teraflops, a tenfold increase in computational capabilities over its predecessor, Deepthought. A committee of campus faculty researchers, each with significant experience in HPC environments and together representing a variety of disciplines, helped the Division of Information Technology (DIT) develop the technical specifications for Deepthought2.

The committee's goal was to design a resource that meets diverse computational needs. True to their vision, the cluster offers about 9,200 CPU based cores for generalized computation, and about 200,000 GPU based cores  for specialized processing (including visualization), and four large memory nodes for data-intensive and serial/threaded applications, such as those used by researchers in genomics, geophysics, etc. In addition, Deepthought2 has a petabyte of directly-attached (Lustre) storage, is interconnected by means of a 56 gigabit InfiniBand network, and connects to campus via a 10 Gb fiber link.

The Division of IT partnered with several colleges to obtain this resource and to upgrade the Cyberinfrastructure Center that houses it. Significant contributors include Jayanth Banavar, Professor and Dean, and Sergei Sukharev, Professor, both in the College of Computer, Mathematical, and Natural Sciences; John Townshend, Professor and Dean, in the College of Behavioral and Social Sciences; and Darryll Pines, Professor and Dean, Yifei Mo, Assistant Professor, Elaine Oran, Professor, and Arnaud Trouvé, Assistant Professor, all in the A. James Clark School of Engineering.

Derek Richardson, Professor of Astronomy, said of the new cluster:

With more than five times the performance and ten times the capacity of the existing cluster, Deepthought2 will allow us to carry out simulations that previously could only be run on national supercomputers, such as high-resolution models of structure formation in the universe or simulations of particles colliding in Saturn's rings. This in turn will bring in new grant money to fund the science enabled by such a powerful local facility.

For further information about the university's high-performance computing resources and access to them, visit http://hpcc.umd.edu or send email to it-hpcc@umd.edu.