CUDA Spotlight: GPU-Accelerated Discovery
Find other great interviews on our CUDA Spotlights page.
This week’s spotlight is on Dr. Axel Kohlmeyer, an expert in high-performance computing with a focus on molecular dynamics (MD) and visualization. Dr. Kohlmeyer is associate director of the Institute for Computational Molecular Science and associate professor of research in the College of Science and Technology at Temple University.
NVIDIA: Axel, when did you start using GPUs?
Axel: John Stone of the University of Illinois is to blame for my involvement in GPUs :). Since about 2003 I have become more and more involved with John's main project, VMD (the molecular visualization program).
A few years ago, John told me about CUDA, a new method to program GPUs that would allow massive speedups. Initially, I had been hesitant since many previous accelerators (like transputers, for example) had promised a large performance increase, but the effort to convert applications to use them had often been voided by the performance increase of conventional processors. However, the GPU-accelerated interactive molecular orbital reconstruction in VMD was showing incredible speedups over what was already a very efficient, multi-threaded, CPU based implementation.
I got started for real with GPUs when I learned about HOOMD (Highly Optimized Object-oriented Many-particle Dynamics), written by Joshua Anderson. It turned out that HOOMD was a good match for a new coarse grain molecular dynamics model being developed in our group. Two colleagues and I implemented a few additionally required features into HOOMD and were impressed by the performance of running MD on a GPU.
NVIDIA: What is exciting in molecular dynamics today?
Axel: Advances in processor speed and parallelization (including CUDA) are allowing us to undertake calculations routinely that were "impossible" just a short time ago. This has spawned a lot of creativity in studying compound systems like large bio-molecules embedded into realistic environments such as membranes. Still, on the fastest supercomputers we are about to reach the boundaries of current algorithms. This has resulted in two interesting trends:
a) Building coarse grain models that realistically describe the behavior of whole molecules without full atomic detail;
b) Using computationally complex models that allow a more realistic description of the atomic interactions without having to use much more time consuming quantum mechanical calculations.
The latter case is particularly interesting for GPU computing, since GPUs "love" computationally-intensive models.
NVIDIA: Temple recently installed a new high-performance computing cluster, based on an NSF grant.
Axel: This is a big win for Temple and a great opportunity in many different ways. Many areas of research are moving towards using high-performance computing tools, thus well-funded and well -supported local facilities are important. And in fields that were traditionally already using HPC, the demands are growing beyond the point where individual research groups can procure, set up and maintain their own resources. I am expecting a lot of synergistic effects from the new machine and I am already working with scientists and students from a variety of departments at Temple to use the new cluster for their research.
NVIDIA: We've read about your involvement in projects to help children learn about science, such as the Nano Dome at the recent Philadelphia Science Festival.
Axel: Science education has become a big problem. There is a lack of well-qualified teachers and class offerings. To become a scientist you have to be a bit crazy, a bit idealistic, and believe that science is more than just a job, that it is "cool."
The "coolness factor" is the driving force behind the Nano Dome, a collaboration with the TUteach project here at Temple, Creative Consultants of Albuquerque, New Mexico, and many others. The Nano Dome is an immersive, interactive, force-feedback computer simulation of nano-particles like carbon nanotubes and C60 fullerene (aka Buckyballs).
In the back of the dome is a powerful GPU-based computer that runs a simulation using a realistic model of atoms and molecules that you interact with using a force feedback controller (a Novint Falcon game controller). This allows you to "feel" how atoms bump into each other, or how mechanics on the nano scale can be very different from a corresponding "macroscale" object.
Since you sit inside the geodesic dome with stereo projection, you see the objects right in front of you. It is easier to experience than to describe, since the idea behind this kind of immersive interaction is to stimulate as many senses as possible, to make a deeper lasting impression. We had a lot of "children of all ages" experience atoms in motion and then step out of the dome and say "cool!"
NVIDIA: As computing becomes more powerful, what does the future hold?
Axel: Computing will become much more pervasive and ubiquitous, up to the point where we will often forget that we are using a computer. However, this also presents a huge challenge: where are we going to find people who are capable and motivated enough to develop future technologies? In many ways, we have begun to think that things "just work" and so we have stopped wondering what makes them "tick".
When I was a child, I loved the feeling of discovery. This is what ultimately drove me to work in an academic environment, where I have the opportunity to discover something new every day and experience the excitement that motivated me as a child. I am concerned that kids these days are missing out on that sense of discovery, the "rush" that you feel when you suddenly can connect the dots and understand what is going on.
- Virtual Atomic Force Microscope using Interactive MD with LAMMPS and VMD: http://www.youtube.com/watch?v=Og9M7xIhW00
- Interactive MD with LAMMPS and VMD:
- The Computational Science Behind Shampoo:
Personal Homepage: http://sites.google.com/site/akohlmey/
Curriculum Vitae: http://sites.google.com/site/akohlmey/curriculum-vitae
ICMS Homepage: http://www.icms.temple.edu/
TUteach Homepage: http://www.temple.edu/cst/tuteach/
VMD Homepage: http://www.ks.uiuc.edu/Research/vmd/
HOOMD Homepage: http://codeblue.umich.edu/hoomd-blue/
LAMMPS Homepage: http://lammps.sandia.gov/
Philadelphia Science Festival Homepage: http://philasciencefestival.org/
Creative Consultants: http://www.creativec.com/
1. Fast Analysis of Molecular Dynamics Trajectories with Graphics Processing Units: Radial Distribution Function Histogramming, B.G.Levine, J.E. Stone, A. Kohlmeyer, J. Comput. Phys. 230(9), 3556-3569 (2011). (Link)
2. Immersive Molecular Visualization and Interactive Modeling with Commodity Hardware, J. E. Stone, A. Kohlmeyer, K. L. Vandivort, and K. Schulten, Lecture Notes in Computer Science, 6454 382-393 (2010). (Link)
3. Ab initio molecular dynamics study of supercritical carbon dioxide including dispersion corrections , S. Balasubramanian, A. Kohlmeyer, M. L. Klein, Journal of Chemical Physics 131, 144506 (2009). (Link)
Location: Institute for Computational Molecular Science Department of Chemistry, College of Science & Technology, Bio-Life Building, Room 113D, Temple University, Philadelphia, PA 19122