Three parallel algorithms for classical molecular dynamics are presented. The first assigns each processor a fixed subset of atoms; the second assigns each a fixed subset of inter-atomic forces to compute; the third assigns each a fixed spatial region. The algorithms are suitable for molecular dynamics models which can be difficult to parallelize efficiently—those with short-range forces where the neighbors of each atom change rapidly. They can be implemented on any distributed-memory parallel machine which allows for message-passing of data between independently executing processors. The algorithms are tested on a standard Lennard-Jones benchmark problem for system sizes ranging from 500 to 100,000,000 atoms on several parallel supercomputers--the nCUBE 2, Intel iPSC/860 and Paragon, and Cray T3D. Comparing the results to the fastest reported vectorized Cray Y-MP and C90 algorithm shows that the current generation of parallel machines is competitive with conventional vector supercomputers even for small problems. For large problems, the spatial algorithm achieves parallel efficiencies of 90% and a 1840-node Intel Paragon performs up to 165 faster than a single Cray C9O processor. Trade-offs between the three algorithms and guidelines for adapting them to more complex molecular dynamics simulations are also discussed.

Fast parallel algorithms for short-range molecular dynamics

We describe a global optimization technique using “basin-hopping” in which the potential energy surface is transformed into a collection of interpenetrating staircases. This method has been designed to exploit the features that recent work suggests must be present in an energy landscape for efficient relaxation to the global minimum. The transformation associates any point in configuration space with the local minimum obtained by a geometry optimization started from that point, effectively removing transition state regions from the problem. However, unlike other methods based upon hypersurface deformation, this transformation does not change the global minimum. The lowest known structures are located for all Lennard-Jones clusters up to 110 atoms, including a number that have never been found before in unbiased searches.

/pdf/global-optimization-by-basin-hopping-and-the-lowest-energy-2s1wxtsj18.pdf

Global Optimization by Basin-Hopping and the Lowest Energy Structures of Lennard-Jones Clusters Containing up to 110 Atoms

We present an implementation of explicit solvent all atom classical molecular dynamics (MD) within the AMBER program package that runs entirely on CUDA-enabled GPUs. First released publicly in April 2010 as part of version 11 of the AMBER MD package and further improved and optimized over the last two years, this implementation supports the three most widely used statistical mechanical ensembles (NVE, NVT, and NPT), uses particle mesh Ewald (PME) for the long-range electrostatics, and runs entirely on CUDA-enabled NVIDIA graphics processing units (GPUs), providing results that are statistically indistinguishable from the traditional CPU version of the software and with performance that exceeds that achievable by the CPU version of AMBER software running on all conventional CPU-based clusters and supercomputers. We briefly discuss three different precision models developed specifically for this work (SPDP, SPFP, and DPDP) and highlight the technical details of the approach as it extends beyond previously reported work [Gotz et al., J. Chem. Theory Comput. 2012, DOI: 10.1021/ct200909j; Le Grand et al., Comp. Phys. Comm. 2013, DOI: 10.1016/j.cpc.2012.09.022].We highlight the substantial improvements in performance that are seen over traditional CPU-only machines and provide validation of our implementation and precision models. We also provide evidence supporting our decision to deprecate the previously described fully single precision (SPSP) model from the latest release of the AMBER software package.

/pdf/routine-microsecond-molecular-dynamics-simulations-with-12m59cn3pd.pdf

Routine Microsecond Molecular Dynamics Simulations with AMBER on GPUs. 2. Explicit Solvent Particle Mesh Ewald.

Molecular dynamics (MD) allows the study of biological and chemical systems at the atomistic level on timescales from femtoseconds to milliseconds. It complements experiment while also offering a way to follow processes difficult to discern with experimental techniques. Numerous software packages exist for conducting MD simulations of which one of the widest used is termed Amber. Here, we outline the most recent developments, since version 9 was released in April 2006, of the Amber and AmberTools MD software packages, referred to here as simply the Amber package. The latest release represents six years of continued development, since version 9, by multiple research groups and the culmination of over 33 years of work beginning with the first version in 1979. The latest release of the Amber package, version 12 released in April 2012, includes a substantial number of important developments in both the scientific and computer science arenas. We present here a condensed vision of what Amber currently supports and where things are likely to head over the coming years. Figure 1 shows the performance in ns/day of the Amber package version 12 on a single-core AMD FX-8120 8-Core 3.6GHz CPU, the Cray XT5 system, and a single GPU GTX680. © 2012 John Wiley & Sons, Ltd.

/pdf/an-overview-of-the-amber-biomolecular-simulation-package-40kdnrsio7.pdf

An overview of the Amber biomolecular simulation package

The present invention concerns polypeptides comprising a variant Fc region. More particularly, the present invention concerns Fc region-containing polypeptides that have altered effector function as a consequence of one or more amino acid modifications in the Fc region thereof.

Polypeptide variants with altered effector function

In 1959, Anfinsen demonstrated that the primary structure (the sequence of amino acids) of a protein can uniquely determine its tertiary structure (three dimensional conformation). This implied that there must be a consistent set of rules for deriving a protein’s tertiary structure from its primary structure. The search for these rules is known as the protein folding problem. Despite many creative attempts, these rules have not been determined (Fasman, 1989). Currently, the primary structures of approximately 40,000 proteins are known. However, only a small percentage of those proteins have known, tertiary structures. A solution to the protein folding problem will make 40,000 more tertiary structures available for immediate study by translating the DNA sequence information in the sequence databases into three-dimensional protein structures. This translation will be indispensable for the analysis of results from the Human Genome Project, de novo protein design, and many other areas of biotechnological research. Finally, an in-depth study of the rules of protein folding should provide vital clues to the protein folding process. The search for these rules is therefore an important objective for theoretical molecular biology.

The Genetic Algorithm and Protein Tertiary Structure Prediction

One embodiment of the present invention sets forth a technique for synchronizing the execution of multiple cooperative thread arrays (CTAs) implementing a parallel algorithm that is mapped onto a graphics processing unit. An array of semaphores provides synchronization status to each CTA, while one designated thread within each CTA provides updated status for the CTA. The designated thread within each participating CTA reports completion of a given computational phase by updating a current semaphore within the array of semaphores. The designated thread then polls the status of the current semaphore until all participating CTAs have reported completion of the current computational phase. After each CTA has completed the current computational phase, all participating CTAs may proceed to the next computational phase.

Method for synchronizing independent cooperative thread arrays running on a graphics processing unit

The present disclosure is directed to parallelization of artificial neural network processing by conditionally synchronizing, among multiple computer processors, either the input or output of individual operations, and by conditionally using either rows or columns of certain matrices used in the operations. The conditional processing may depend upon the relative sizes of the input and output of the specific operations to be performed. For example, if a current layer matrix of values is larger than a next layer matrix of values to be computed, then rows of a weight matrix may be used by the computer processors to compute the next layer matrix. If the current layer matrix is smaller than the next layer matrix, then columns of conditional parallel processing in fully-connected neural networksthe weight matrix may be used by the computer processors to compute the next layer matrix.

Conditional parallel processing in fully-connected neural networks

One embodiment of the present invention sets forth a technique for efficiently performing a radix sort operation on a graphics processing unit (GPU). The radix sort operation is conducted on an input list of data using one or more passes of a series of three processing phases. In each processing phase, thread groups are each associated with one segment of input data. In the first phase, occurrences of each radix symbol are counted and stored in a list of counters. In the second phase, the list of counters is processed by a parallel prefix sum operation to generate a list of offsets. In the third phase, the list of offsets is used to perform re-ordering on the list of data, according to the current radix symbol. To maintain sort stability, the one or more passes proceed from least significant data to most significant data in the sort key.

Reordering data using a series of offsets

A service provider system may implement ECC-like features when executing computations on GPUs that do not include sufficient error detection and recovery for computations that are sensitive to bit errors. During execution of critical computations on behalf of customers, the system may automatically instrument program instructions received from the customers to cause each computation to be executed using multiple sets of hardware resources (e.g., different host machines, processor cores, or internal hardware resources). The service may provide APIs with which customers may instrument their code for execution using redundant resource instances, or specify parameters for applying the ECC-like features. The service or customer may instrument code to perform (or cause the system to perform) checkpointing operations at particular points in the code, and to compare intermediate results produced by different hardware resources. If the intermediate results do not match, the computation may be restarted from a checkpointed state.

Scott M. Le Grand

Papers

The Genetic Algorithm and Protein Tertiary Structure Prediction

Method for synchronizing independent cooperative thread arrays running on a graphics processing unit

Conditional parallel processing in fully-connected neural networks

Reordering data using a series of offsets

Automated error detection and recovery for GPU computations in a service environment