Overview
Molecular Dynamics (MD) simulations of complex systems rely heavily on large-scale computations. Graphics processing units (GPUs) with hundreds of arithmetic units can process data in a highly parallel manner, and thus provide significant acceleration over central processing units (CPUs). In recent years, GPU-accelerated MD simulations have emerged to be a powerful approach to investigate conformational dynamics of complex chemical and biological systems, and can provide significant acceleration compared to CPUs. For instance, the GPU acceleration has been implemented in all mainstream MD software including GROMACS, NAMD, and AMBER.
The goal of this workshop is to gather major developers of GPU-accelerated MD simulation software to discuss the recent progress in this field. The topics will focus particularly on the algorithmic advances that can greatly benefit from parallel architecture of GPUs, and the large-scale applications of GPU-accelerated MD simulations.
The following lectures in this workshop are sponsored by IAS:
|
Lecture 1: GPU-acceleration and New Streaming Algorithms for Molecular Simulations |
|
|
|
Date |
: |
12 Dec 2018 (Wed) |
|
Time |
: |
9:10 – 9:50 am |
|
Speaker |
: |
Prof. Erik LINDAHL, Stockholm University and KTH Royal Institute of Technology |
|
|
Abstract
Molecular dynamics simulation and many other scientific applications have seen great speedup on modern accelerators, in particular graphics processors, but this is part of a larger trend where all modern processors have become extremely parallel. In this talk, the speaker will start with an overview of this development and illustrate how he has optimized the GROMACS molecular dynamics code for a range of different accelerators, and how this has forced him to revisit and change many of the fundamental algorithms developed for molecular simulations in the last 40 years. It turns out that these ways of rewriting algorithms are usually very beneficial also for traditional CPUs where he was convinced his previous code was already optimal. He will illustrate what this enables him to do with a handful of application examples, and also focus on several important lessons that teach him about the future of scientific computing: How fast can we expect simulations to get? How to program new computers to do this? How should scientists design their simulation studies to fully use the power of accelerators efficiently for normal medium-to-small size systems? Finally, he will illustrate how merely learning to think in terms of accelerator computing might make it possible for scientists to transform adjacent research areas and help science advance significantly.
About the speaker
Prof. Erik Lindahl received his PhD in Theoretical Biophysics from KTH Royal Institute of Technology in Stockholm in 2001. He then joined the Stockholm University in 2004 and is currently the Professor of Biophysics.
Prof. Lindahl received the Sven and Ebba-Christina Hagberg Prize by the Royal Swedish Academy of Sciences in 2005. He was elected a Member of the Young Academy of Sweden in 2011.
|
Lecture 2: GPU Acceleration of Biomolecular Simulation and Visualization in NAMD and VMD |
|
|
|
Date |
: |
13 Dec 2018 (Thu) |
|
Time |
: |
9:00 – 9:40 am |
|
Speaker |
: |
Prof. Emad TAJKHORSHID, University of Illinois at Urbana-Champaign |
|
|
Abstract
Modern GPUs have impacted scientific computing in major ways. Nearly all developers of major simulation and visualization software packages have been investing in porting algorithms to GPUs so that they can take advantage of the unparalleled speed provided by them. The US National Institutes of Health Center for Macromolecular Modeling and Bioinformatics at Illinois continues to enhance and expand the scope of its two flagship software packages, NAMD and VMD, the molecular dynamics code and the visualization software, respectively. In this talk, the speaker will report on his progress on the enhancement of features and capabilities of these software packages over the recent years, specifically in the context of integration of GPU capabilities. Examples include improvement of performance of NAMD, including free energy calculation algorithms, as well as several VMD enhanced capabilities in real-time rendering and handling very large molecular systems. He will also discuss several examples of unique applications enabled by the computational power of GPUs, including simulation of the first billion-atom bio-molecular system, visualization of very large molecular assemblies, and remote visualization capabilities in VMD.
About the speaker
Prof. Emad Tajkhorshid received his PhD in Medicinal Chemistry at University of Tehran (1995) and a second PhD in Molecular Biophysics from Heidelberg University (2001). He joined the University of Illinois at Urbana-Champaign (UIUC) in 2004 and is currently the J. Woodland Hastings Endowed Chair in Biochemistry, Professor of Bioengineering, Biophysics, and Quantitative Biology.
Prof. Tajkhorshid received the Faculty Excellence Award by UIUC (2016) and was named as the University of Illinois Scholar (2015).
About the Workshop
For more information, please refer to the workshop website at http://compbio.ust.hk/GPU_Conference/.
|
