Paper on GPU-accelerated molecular modeling is “Most Downloaded” for journal
The Journal of Molecular Graphics and Modelling announced that “GPU-Accelerated Molecular Modeling Coming of Age,” co-authored by three Illinois researchers and a Stanford colleague, is the journal’s #1 most-downloaded research paper as of October 27, 2010. The research was presented through the University of Illinois CUDA Center of Excellence, headquartered at the Coordinated Science Laboratory.
The invited paper was written by researchers John Stone and David J. Hardy (both of UIUC Beckman Institute), and Prof. Klaus Schulten (UIUC Dept. of Physics and Beckman Institute), along with Stanford’s Ivan S. Ufimtsev. They review the traditional role that graphics processing units (GPUs) have played in molecular modeling, where they were once used solely for visualization of structures and animation of trajectories resulting from molecular dynamics simulations. But modern GPUs have evolved into fully programmable, massively parallel co-processors that can accelerate many scientific computations, providing about one to two orders of magnitude speedup over CPU code. The authors survey the development of molecular modeling algorithms that leverage GPU speedups, the advances already made, and the continuing evolution of GPU technology, which promises to become even more useful to molecular modeling.
The paper was published in the September 2010 issue of JMGM. According to co-author John Stone, "In the months since we submitted our paper, we have seen a tremendous amount of activity in GPU computing for molecular modeling, as reflected in presentations in a special session at the Fall meeting of the American Chemical Society, and at the September GPU Technology Conference. And about one week ago, a GPU-accelerated supercomputer, also intended for biomolecular modeling, was named the fastest machine in the world. We're happy that our paper is getting so much attention as part of the widespread community interest in GPU computing for molecular modeling."
The authors also point the way to the future of GPUs in molecular modeling, as they posit that hardware acceleration with commodity GPUs will benefit the overall computational biology community by bringing teraflops performance to desktop workstations -- in some cases changing what were formerly batch-mode computational jobs into interactive tasks.
This research was supported by NIH grant P41-RR05969. Performance experiments were made possible by a generous hardware donation by NVIDIA. Ivan Ufimtsev acknowledges an NVIDIA fellowship and NSF grant CHE-06-26354.