This collaborative pilot project between the San Diego Supercomputer Center at the University of California San Diego, the Quantum Theory Project at the University of Florida and industrial partners NVIDIA Inc. is focused on developing innovative, comprehensive open source software element libraries for accelerating condensed phase Molecular Dynamics (MD) simulations of biomolecules using Graphics Processing Units (GPU). By porting MD techniques to GPUs this project is enabling users to both attain substantial increases in their own local calculations without the need for substantial investment in hardware or infrastructure, and to make effective use of GPU acceleration provided by new machines within the NSF supercomputing centers. The software elements being developed and distributed both within the AMBER MD package and as open source libraries are providing critical software infrastructure in support of transformative research in the fields of chemistry, life science, materials science, environmental and renewable energy research.
The software elements being created in this project have very broad impact. For example, the integration of single and multi-GPU acceleration within the AMBER software alone benefits a very large and established national and international user base. Over 8,000 downloads of the AMBER Tools package from unique IP addresses and more than 500 sites which use the AMBER MD engine testify to the scope of the community of researchers this work impacts. Additionally the open source GPU MD acceleration libraries being produced provide broad impact across multiple domains while outreach workshops are helping to train the next generation of scientists not just in the use and potential benefits of GPU MD acceleration libraries but also in modern MD simulation techniques.
This award is co-funded by the Office of Cyberinfrastructure, the Division of Chemistry and the Office of Multidisciplinary Activites of the Directorate of Mathematical Sciences.
Project Title: Collaborative Research SI2-SSE: Comprehensive Sustained Innovation in Acceleration of Molecular Dynamics Simulation and Analysis on Graphical Processing. PI: Ross Walker Awardee: University of California San Diego Award Number: 1047875 The project was a one-year collaborative pilot project to extend the use of GPGPUs (General Purpose Graphical Processing Units, a.ka. graphic cards) to the field of molecular dynamics, in particular to the program Amber. GPGPUs represent an extensive step in terms of how much computation can be done in a given time, within a given budget and within certain power consumption. They also lower the barrier to entry into the field, by creating a level playing field at a cost that all research groups can easily bare. We focused on the program Amber, a widely used set of tools to perform Molecular Dynamics of biological systems. With our current implementation, we can achieve performances on 8 GPUS that cannot be reached even with the fastest conventional supercomputers available in the US at this time. This work has been conducted in collaboration with NVIDIA, providing an example of Academic/Industry collaborations. We have also collaborated with a different group funded by NSF, that of Prof. Todd Martinez (Stanford University), to perform quantum calculations using GPUs, coupled to the program Amber. The work has also been featured in a number of press releases in both the scientific and non-scientific press including a feature length article in NVISION magazine, and press articles in PC Magazine, Hardware Zone and HPC Wire (http://tinyurl.com/5wnndds) to name just a few. The work was also featured as one of the three key science applications highlighted by the NVIDIA CEO in his keynote speech at the 2010 Graphics Technology Conference. A workshop on Molecular dynamics, including examples from our SSE work, was run in May 2011 at the Barcelona Supercomputer Center in Barcelona, Spain. This was jointly funded by the Barcelona Supercomputer Center and NVIDIA and was significantly oversubscribed receiving over 200 applications for just 50 spaces in the first week. We ran a second workshop in Shanghai, China in August 2011 that was jointly funded by the East China Normal University and NVIDIA. The software produced as part of this work is now being used in over 800 labs world wide with citations running in excess of 1,000 per year. This software is used across all domains involving an understanding of enzyme activity. Examples include anti-cancer research, next generation drug design and biocatalysis for renewable energy.
