- Algorithms & method for accelerating molecular dynamics
- Parallelization and acceleration of molecular dyanmics on modern high performance computing architectures
- High performance computing, accelerator architectures, GPU computing
Molecular dynamics simulations (MD) offer a great advantage compared to laboratory experiments by giving unique insights into dynamic processes in biomolecules with an exceptional atomistic detail. However, one of the major bottlenecks in MD studies is sampling. In practice, this translates to computational experiments taking as long as weeks or months. My work aims to contribute to the field with methods and algorithms that improve the efficiency of biomolecular simulations by focusing on the following two aspects:
- Sampling the same amount in less time by speeding up simulations. We approach this computational problem by developing algorithms for modern parallel architectures. In our recent work we developed algorithms targeting modern processor architectures and an efficient parallelization scheme for emerging heterogeneous high performance computing platforms. Through this work we managed to substantially improve absolute performance and scalability of MD simulations using the state-of-the art GROMACS package. Our work enables thousands of users worldwide to achieve 2-5x higher absolute simulation performance.
- Sampling more in the same amount of time by improving ensemble methods. We worked on using parallel simulations which exchange information to increase the efficiency of calculating free energy differences.
The goal is to combine the developed methods and algorithms with the advantages of the distributed computing platform Copernicus developed in our group.
Szilárd Páll, Berk Hess, A flexible algorithm for calculating pair interactions on SIMD architectures, Computer Physics Communications, Available online 13 June 2013, ISSN 0010-4655
Sander Pronk, Szilárd Páll, Roland Schulz, Per Larsson, Pär Bjelkmar, Rossen Apostolov, Michael R. Shirts, Jeremy C. Smith, Peter M. Kasson, David van der Spoel, Berk Hess, Erik Lindahl: GROMACS 4.5: a high-throughput and highly parallel open source molecular simulation toolkit. Bioinformatics 29(7): 845-854 (2013)
Wolfgang Schreiner, Karoly Bosa, Andreas Langegger, Thomas Leitner, Bernhard Moser, Szilard Pall, Volkmar Wieser, Wolfram Wöß . Parallel, Distributed, and Grid Computing. In: Hagenberg Research, B. Buchberger, M. Affenzeller, A. Ferscha, M. Haller, T. Jebelean, E. P. Klement, P. Paule, G. Pomberger, W. Schreiner, R. Stubenrauch, R. Wagner, G. Weiss, W. Windsteiger (ed.), Chapter VII, pp. 333-378. 2009. Springer, Berlin, ISBN 78-3-642-02126-8.
Education & Background
2009 PhD in computational biophysics – started at Stockholm University then moved to KTH Royal Institute of Technology in 2011.
2008-2009 software engineer and member of the scientific staff at Software Competence Center Hagenberg
Worked as R&D software engineer on various problems in the field of computational intelligence related to GPU acceleration and machine learning.
Master’s thesis: “GPU Computing Approach for Parallelizing Support Vector Machine Classification”
2007 BSc in computer science from Babeș-Bolyai University, Cluj-Napoca, Romania
Diploma work: “Text Categorization-based Spam Filtering”