Scientific American Article Features Work of CS Faculty

Publish Date: 10/26/2009

Scientific American’s article “ New Software Could Smooth Supercomputing Speed Bumps” looks at how researchers are turning to the Open Computing Language to enable software writers to create applications that can run on both GPUs and CPUs with minimal modifications. Within the article, the research of Virginia Tech Computer Science faculty members Dr. Wu Feng and Dr. Alexey Onufriev is highlighted. Feng and Onufriev’s research demonstrates how a computer equipped with both a CPU and GPU drastically decreases the time it takes to compute and visualize molecular simulations – reducing the computing time from over 22 hours to less than a minute!

Molecular simulations at full atomic resolution are, arguably, one of the most accurate computational tools available to researchers that study molecular function at the basic level. As one can expect, such simulations are extremely demanding computationally.  Approaches to speed-up these computations can generally be subdivided into two very broad categories: (1) those that seek to gain speed by making computationally effective approximations to the underlying physical model and (2) those that do not affect the accuracy of the physical model but strive to accelerate the computation at the software or hardware levels.

The work demonstrates that two emerging methods from these different categories can be combined to accelerate typical "large scale" biomolecular computations by several orders of magnitude. Specifically, the Hierarchical Charge Partitioning approximation, developed by Ramu Anandakrishnan (graduate student in Alexey Onufriev's lab), was adapted to the graphics processing unit (GPU) by Tom Scogland (graduate student in Wu Feng's lab) to speed-up computations of electrostatic properties of macromolecules. A calculation of electrostatic potential produced by a giant viral capsid now takes 35 seconds, as opposed to 22 hours.  The idea to test if these types of electrostatic computations can be accelerated dramatically by the GPU came from Physics graduate student Andrew T. Fenley (also in Onufriev's lab), who, along with a former CS graduate student John C. Gordon, developed the original electrostatic algorithm.