With this award from the Chemistry Research Instrumentation and Facilities: Multi-user (CRIF:MU) program, Professor Michael Marletta and colleagues Phillip Geissler, Martin Head-Gordon, William Lester, Jr and K. Birgitta Whaley from University of California at Berkeley will acquire a multi-teraflop cluster of graphical processing units (GPUs) that will be used to build a new generation of advanced algorithms. The proposal is aimed at enhancing research training and education at all levels, especially in areas of study such as (a) nano-scale systems, (b) development and application of molecular simulation techniques, (c) large scale simulations, (d) development and applications of theoretical tools to study a wide range of problems in statistical mechanics, (e) elucidation of mechanisms of biomolecular communication, and (f) development of theory, algorithms and software to describe the electronic structure of molecules.
A computer cluster is a group of linked processors that work in concert to achieve vastly more computational power that the individual computers. Such a "computer network" serves as a development environment for new theoretical codes and algorithms, provides state-of-the-art graphics and visualization facilities, and supports research in state-of-the-art applications of parallel processing. The computing resource will permit research, research training, and research outreach opportunities the collaborating theory groups at Howard University and Jackson State University. A portion of the cluster will be reserved for use by young scientists, who will be trained via workshops and courses.
Under support from the Chemistry Research Instrumentation and Facilities (CRIF) program, a shared computer facility containing both conventional computer and unconventional graphical processor units (GPU's) was purchased for use by the theoretical chemistry faculty at the University of California, Berkeley. The machine was purchased in two phases over three years, to minimize the rapid obsolescence of computer hardware, as well to permit early experimentation with the GPU capability. The machine presently has a total of 180 user accounts and is used to capacity. The primary users are spread across 7 theoretical and computational research groups, who have reported over 10 publications using the facility during its last year alone. The topics ranged from computing accurate intermolecular interaction energies to the testing and evaluation of newly developed electronic structure methods, to the simulation of amorphous ices, to the exploration of decoherence behavior in quantum computing, to structural heterogeneity within protein cores. A general finding was that the barrier to developing GPU-specific code was high, but such codes offered great computational advantages after they were production ready. Since computational methods touch more and more areas of chemistry, outreach to support the training of non-specialist graduate students in the use of computational methods was performed, which involved providing over 20 dedicated user accounts on the system in the current semester. Additionally, an external collaboration with a researcher at a leading historically black college and university was established as part of this project.
