We describe the algorithm of employing multi-GPU power on the basis of Message Passing Interface (MPI) domain decomposition in a molecular dynamics code, GALAMOST, which is designed for the ...coarse-grained simulation of soft matters. The code of multi-GPU version is developed based on our previous single-GPU version. In multi-GPU runs, one GPU takes charge of one domain and runs single-GPU code path. The communication between neighbouring domains takes a similar algorithm of CPU-based code of LAMMPS, but is optimised specifically for GPUs. We employ a memory-saving design which can enlarge maximum system size at the same device condition. An optimisation algorithm is employed to prolong the update period of neighbour list. We demonstrate good performance of multi-GPU runs on the simulation of Lennard-Jones liquid, dissipative particle dynamics liquid, polymer and nanoparticle composite, and two-patch particles on workstation. A good scaling of many nodes on cluster for two-patch particles is presented.
Recently, most cloud services use Docker container environment to provide their services. However, there are no researches to evaluate the performance of communication libraries for multi-GPU based ...distributed deep learning in a Docker container environment. In this paper, we propose an efficient communication architecture for multi-GPU based deep learning in a Docker container environment by evaluating the performances of various communication libraries. We compare the performances of the parameter server architecture and the All-reduce architecture, which are typical distributed deep learning architectures. Further, we analyze the performances of two separate multi-GPU resource allocation policies ― allocating a single GPU to each Docker container and allocating multiple GPUs to each Docker container. We also experiment with the scalability of collective communication by increasing the number of GPUs from one to four. Through experiments, we compare OpenMPI and MPICH, which are representative open source MPI libraries, and NCCL, which is NVIDIA’s collective communication library for the multi-GPU setting. In the parameter server architecture, we show that using CUDA-aware OpenMPI with multi-GPU per Docker container environment reduces communication latency by up to 75%. Also, we show that using NCCL in All-reduce architecture reduces communication latency by up to 93% compared to other libraries.
The airframe/engine integrated configurations with dorsal intake can effectively decouple the inlet from the lifting surface, leading to an excellent aerodynamic performance at a small angle of ...attack at hypersonic speeds. For a comprehensive and systematic investigation of the dorsal intake configuration, it is necessary to compare it with a ventral intake configuration of equal scale. However, most of the present assessments for dorsal intake configurations focus on the fundamental aerodynamic characteristics with a lack of stability and control characteristics. Therefore, the computational fluid dynamics numerical simulation based on multi-GPU parallel computing is introduced for comparing the basic static stability and control characteristics between the dorsal and ventral intake configurations. The results show that the dorsal intake configuration has superior three-channel static stability at hypersonic speeds than the ventral intake configuration. However, this also results in worse control characteristics, which may be remedied by relaxing the static stability margin or installing canards. Moreover, the selection of intake configuration only interferes with the coupling of the pitch channel and does not affect the coupling of the yaw and roll channels.
•Multi-GPU accelerated topology optimization using aggregation-based AMG.•Low setup cost and device memory requirements using efficient interpolation.•Multiple GPUs accelerate computing and increase ...the device memory available.•Significant speedups in comparison with efficient multi-core computing.•The use of mixed-precision increases performance.
This work presents a parallel implementation of density-based topology optimization using distributed GPU computing systems. The use of multiple GPU devices allows us accelerating the computing process and increasing the device memory available for GPU computing. This increment of device memory enables us to address large models that commonly do not fit into one GPU device. The most modern scientific computers incorporate these devices to design energy-efficient, low-cost, and high-computing power systems. However, we should adopt the proper techniques to take advantage of the computational resources of such high-performance many-core computing systems. It is well-known that the bottleneck of density-based topology optimization is the solving of the linear elasticity problem using Finite Element Analysis (FEA) during the topology optimization iterations. We solve the linear system of equations obtained from FEA using a distributed conjugate gradient solver preconditioned by a smooth aggregation-based algebraic multigrid (AMG) using GPU computing with multiple devices. The use of aggregation-based AMG reduces memory requirements and improves the efficiency of the interpolation operation. This fact is rewarding for GPU computing. We evaluate the performance and scalability of the distributed GPU system using structured and unstructured meshes. We also test the performance using different 3D finite elements and relaxing operators. Besides, we evaluate the use of numerical approaches to increase the topology optimization performance. Finally, we present a comparison between the many-core computing instance and one efficient multi-core implementation to highlight the advantages of using GPU computing in large-scale density-based topology optimization problems.
•We propose and implement a bonded and short-range scalable molecular dynamics for distributed Multi-GPU environments.•Our approach makes the calculations of all forces completely in GPUs.•The ...algorithms, CUDA kernels and MPI communication are designed to improve scalability and runtime performance.•We test its scalability compared to NAMD by simulating on Amazon Web Services using both K80 and V100 GPUs.•We obtain linear horizontal scalability in multiple scenarios, including simulations of 1 billion atom molecules.
GPUs have enabled a drastic change to computing environments, making massively parallel computing possible. Molecular dynamics is a perfect candidate problem for massively parallel computing, but to date it has not taken full advantage of multi-GPU environments due to the difficulty of partitioning molecular dynamics problems and exchanging problem data among compute nodes. These difficulties restrict the use of GPUs to only some of the computations in a full molecular dynamics problem, and hence prevent scalability beyond just a few GPUs. This work presents a scalable parallelization solution for the bonded and short-range forces present in a molecular dynamics problem. Together with existing solutions for long-range forces, it enables highly scalable, parallel molecular dynamics on multi-GPU computing environments. Specifically, the proposed solution divides the molecular volume into independent parts assigned to different GPUs, but it maintains a global bond structure that is efficiently exchanged when atoms move across GPUs. We demonstrate close-to-linear speedup of the proposed solution, simulating the dynamics of gigamolecules with 1 billion atoms on a computing environment with 96 GPUs, and obtaining superior performance to the well known molecular dynamics simulator NAMD.
We present a novel approach for volume exploration that is versatile yet effective in isolating semantic structures in both noisy and clean data. Specifically, we describe a hierarchical active ...contours approach based on Bhattacharyya gradient flow which is easier to control, robust to noise, and can incorporate various types of statistical information to drive an edge-agnostic exploration process. To facilitate a time-bound user-driven volume exploration process that is applicable to a wide variety of data sources, we present an efficient multi-GPU implementation that (1) is approximately 400 times faster than a single thread CPU implementation, (2) allows hierarchical exploration of 2D and 3D images, (3) supports customization through multidimensional attribute spaces, and (4) is applicable to a variety of data sources and semantic structures. The exploration system follows a 2-step process. It first applies active contours to isolate semantically meaningful subsets of the volume. It then applies transfer functions to the isolated regions locally to produce clear and clutter-free visualizations. We show the effectiveness of our approach in isolating and visualizing structures-of-interest without needing any specialized segmentation methods on a variety of data sources, including 3D optical microscopy, multi-channel optical volumes, abdominal and chest CT, micro-CT, MRI, simulation, and synthetic data. We also gathered feedback from a medical trainee regarding the usefulness of our approach and discussion on potential applications in clinical workflows.
Multi GPU systems achieve performance improvement through scaling out to meet the ever-increasing computation demand brought about by increasingly complex algorithms and the continuously increasing ...data in artificial intelligence. The interconnection bandwidth between processors, as well as topologies of systems are the key factors that determine the performance of multi-GPU systems. In traditional PCIe-based multi-GPU systems, the PCIe bandwidth is the bottleneck that limits system performance. GPU-oriented high speed interconnection technologies become an effective method to solve the bandwidth limitation problem of multi-GPU systems at present. This article first introduces the PCIe interconnection technology and the typical topologies used in traditional multi-GPU systems. Then taking Nvidia NVLink, AMD Infinity Fabric Link, Intel Xe Link, and Biren Technology BLink as examples, GPU-oriented high speed interconnection technologies and topologies of representative GPU vendors at home and abroad are reviewe
In this paper, a Weakly Compressible Smoothed Particle Hydrodynamics (WCSPH) framework is presented utilizing the parallel architecture of single- and multi-GPU (Graphic Processing Unit) platforms. ...The program is developed for water entry simulations where an efficient potential based contact force is introduced to tackle the interaction between fluid and solid particles. The single-GPU SPH scheme is implemented with a series of optimization to achieve high performance. To go beyond the memory limitation of single GPU, the scheme is further extended to multi-GPU platform basing on an improved 3D domain decomposition and inter-node data communication strategy. A typical benchmark test of wedge entry is investigated in varied dimensions and scales to validate the accuracy and efficiency of the program. The results of 2D and 3D benchmark tests manifest great consistency with the experiment and better accuracy than other numerical models. The performance of the single-GPU code is assessed by comparing with serial and parallel CPU codes. The improvement of the domain decomposition strategy is verified, and a study on the scalability and efficiency of the multi-GPU code is carried out as well by simulating tests with varied scales in different amount of GPUs. Lastly, the single- and multi-GPU codes are further compared with existing state-of-the-art SPH parallel frameworks for a comprehensive assessment.
In information systems, there may exist multiple different types of attributes like categorical attributes, numerical attributes, set-valued attributes, interval-valued attributes, missing ...attributes, etc. Such information systems are called as composite information systems. To process such attributes with rough set theory, composite rough set model and corresponding matrix methods were introduced in our previous study. Calculation of rough set approximations of a concept is the key step for rule acquisition and attribute reduction in rough set based methods. To accelerate the computation process of rough set approximations, this paper first presents the boolean matrix representation of the lower and upper approximations in the composite information system, then designs a parallel method for computing approximations based on matrix, and implements it on Multi-GPU. The experiments on data sets from UCI and user-defined data sets show that the proposed method can accelerate the computation process efficiently. The Multi-GPU implementation achieves up to a speedup of 334.9 over the CPU implementation.