Search

Paper describes the program implementation of a conservative semi-Lagrangian method of 5th order for the graphical processor units (GPUs). Specific approaches to the parallelization using the CUDA architecture are discussed. The developed code is verified using the problem of a collisionless evolution of contact discontinuity. [ABSTRACT FROM AUTHOR]

The technology of adaptation of the HyCFS numerical code, which was originally developed for supercomputers with graphical processor units (GPUs), to various computational platforms, such as conventional CPU-based systems and new supercomputers based on the Intel Xeon Phi co-processors is developed. The main idea of adaptation is to use OpenMP threads instead of CUDA threads. This approach provides a possibility of using a unified source code for different platforms. [ABSTRACT FROM AUTHOR]

Two-dimensional liquid convection problem is implemented using parallel computations based on CUDA technology. The Jacobi method and the conjugate gradient method are used to solve the systems of linear equations (SLE) with sparse matrices, obtained by the initial equations approximation. The calculation time on the central processor of the PC is compared with the time when using GPUs, the calculations acceleration with the unknowns number increase is estimated. [ABSTRACT FROM AUTHOR]

The processes of seismic waves propagation generated by the non-explosive electromagnetic pulse source "Yenisei" are under investigation. Rheological properties of the layered medium including fractured interlayers are taken into account. For numerical implementation, a computational algorithm based on the space-variable two-cyclic splitting method in combination with monotone finite-difference schemes is developed. Parallel software for the analysis of the fields of velocities and stresses in layered medium is designed. Parallelisation of computations is performed by the CUDA technology for supercomputers with graphical accelerators. The results of numerical computations of seismic waves propagation are shown. [ABSTRACT FROM AUTHOR]

Research is focused on parallelization of code in C++ language, using TBB library, OpenMP directives and CUDA technology. There are also comparison of results include. Thanks results we can recommend the best technology for optimization of source code. We present experimental results of computational efficiency of several types of parallelization techniques including GPU execution units. [ABSTRACT FROM AUTHOR]

Parallel computations based on CUDA technology are applied for solving two-dimensional problems. Heat distribution and velocity field calculation in a liquid is considered. Various iterative methods are used to solve simultaneous linear equations with sparse matrices, obtained by the initial equations approximation. The efficiency of using GPUs to accelerate calculations is estimated, which grows with the unknowns number increase. [ABSTRACT FROM AUTHOR]

The paper presents approaches to implementation of solution of discrete dynamical system of mutually repelling particles. Two platforms: a single-thread JAVA process and parallelized CUDA C solution, are employed for the dynamical simulation. Qualities of both platforms are discussed and explained as their performance when solving two proposed interaction laws is compared. [ABSTRACT FROM AUTHOR]

CUDA EGS was CUDA implementation to simulate transport photon in a material based on Monte Carlo algorithm for X-ray imaging. The objective of this study was to investigate the effect of inhomogeneities in inhomogeneity phantom for small field dosimetry (1×1, 2×2, 3×3, 4×4 and 5×5 cm2). Two phantoms, homogeneity and inhomogeneity phantom were used. The interaction in homogeneity and inhomogeneity phantom was dominated by Compton interaction and multiple scattering. The CUDA EGS can represent the inhomogeneity effect in small field dosimetry by combining the grayscale curve between homogeneity and inhomogeneity phantom. The grayscale curve in inhomogeneity phantom is not asymmetric because of the existence of different material in phantom. [ABSTRACT FROM AUTHOR]

The research considers the problem of constructing the reachability sets of non-linear dynamical system in n-dimensional Euclidean space on the fixed time interval. The approximate solution methods of the reachability sets constructing are considered in this research as well as the accuracy estimation for this methods is given. The research contains the computational experiments on computer modeling of described reachability sets constructing methods, which use the algorithms implemented for two computation technologies CPU as well as GPU (using CUDA technology). In this research the description and comparison of approaches to the computer modeling of the problem are given. Furthermore, the CPU-based computer modeling result comparison with the result obtained on GPU based on CUDA technology are presented. Besides, this research discusses some the side issues appeared during computer modeling, the issues raised during the computer algorithms implementation, as well as the ways to eliminate these issues or reduce their impact. [ABSTRACT FROM AUTHOR]

We present a parallel mean shift algorithm running on CUDA and its possible application in segmentation of multispectral images. The aim of this paper is to present a method of analyzing highly noised multispectral images of various objects, so that important features are enhanced and easier to identify. The algorithm finds applications in analysis of multispectral images of eyes so that certain features visible only in specific wavelengths are made clearly visible despite high level of noise, for which processing time is very long. [ABSTRACT FROM AUTHOR]

In the paper the semi-Lagrangian method is considered in the context of its implementation with the CUDA technology. We have scrutinized the bottleneck of our sequential algorithm; its parallel versions are studied in detail; and the main reason of poor CUDA performance is clarified. As the result, we revise the computation of partial integrals in order to improve the efficiency of the algorithm. Numerical experiments demonstrate good CUDA performance of the revised version of the algorithm. [ABSTRACT FROM AUTHOR]

In this work, we present an optimization strategy for implementing the BiCGStab iterative solver on graphic processing units (GPU) for computing incompressible viscous flows governed by the unsteady Navier-Stokes (N-S) equations on a CUDA platform. A recently developed ψ-ν formulation is used to discretize the biharmonic form of the N-S equation and we obtain remarkable speed up of 40 times on finer grids for the lid-driven square cavity flow. The GPU implementation enabled us to compute the flow in extremely finer grids and very small scales were resolved with remarkable accuracy. [ABSTRACT FROM AUTHOR]

This paper describes recent progress towards porting a Unified Flow Solver (UFS) to heterogeneous parallel computing. The main challenge of porting UFS to graphics processing units (GPUs) comes from the dynamically adapted mesh, which causes irregular data access. We describe the implementation of CUDA kernels for three modules in UFS: the direct Boltzmann solver using discrete velocity method (DVM), the DSMC module, and the Lattice Boltzmann Method (LBM) solver, all using octree Cartesian mesh with adaptive Mesh Refinement (AMR). Double digit speedup on single GPU and good scaling for multi-GPU has been demonstrated. [ABSTRACT FROM AUTHOR]

The minimum vertex cover problem is one of the fundamental problems in graph theory and is known to be NPhard. For data mining in large-scale structured systems, we proposes a GPU algorithm for the minimum vertex cover problem. The algorithm is designed to derive sufficient parallelism of the problem for the GPU architecture and also to arrange data on the device memory for efficient coalesced accessing. Through the experimental evaluation, we demonstrate that our GPU algorithm is quite faster than CPU programs and the speedup becomes much evident when the graph size is enlarged. [ABSTRACT FROM AUTHOR]

In the nondestructive evaluation (NDE) of steel plates, besides the detection of cracks, what is also important is their characterization. Characterization is necessary for determining whether any discovered crack demands withdrawal of the part from service. In eddy current crack identification the response of a part to a coil is compared to the response without a crack. When different, the presence of the crack is flagged. But to characterize it, the computed response from eddy current analysis with a crack described by parameters is optimized to match measurements. This is heavy computation. Recently, Graphical Processing Unit (GPU) computing has had great success in many very large numerical computations. In this work we discuss the often undiscussed GPU memory limitation in Finite Element Optimization. In GPU computing the memory of the NVIDIA GPU is limited (4 GB on a PC today). This paper assesses the memory limits in terms of matrix size in light of the various ways to store a large matrix in order to overcome these limits. We revive old element-by-element finite element solvers from the early 1980s of working on a highly memory limited PC 282 to launch a GA kernel on thousands of CUDA threads exploiting the NVIDIA GPU architecture. [ABSTRACT FROM AUTHOR]

We previously reported on an electrical conductivity imaging instrument developed for measurements on polycrystalline diamond cutters. These cylindrical cutters for oil and gas drilling feature a thick polycrystalline diamond layer on a tungsten carbide substrate. The instrument uses electrical impedance tomography to profile the conductivity in the diamond table. Conductivity images must be acquired quickly, on the order of 5 sec per cutter, to be useful in the manufacturing process. This paper reports on successful efforts to optimize the conductivity reconstruction routine, porting major portions of it to NVIDIA GPUs, including a custom CUDA kernel for Jacobian computation. [ABSTRACT FROM AUTHOR]

A software package is developed to perform explicit time domain finite element simulations of ultrasonic propagation on the graphical processing unit, using Nvidia's CUDA. Of critical importance for this problem is the arrangement of nodes in memory, allowing data to be loaded efficiently and minimising communication between the independently executed blocks of threads. The initial stage of memory arrangement is partitioning the mesh; both a well established 'greedy' partitioner and a new, more efficient 'aligned' partitioner are investigated. A method is then developed to efficiently arrange the memory within each partition. The technique is compared to a commercial CPU equivalent, demonstrating an overall speedup of at least 100 for a non-destructive testing weld model. [ABSTRACT FROM AUTHOR]

A numerical model of an ocean global circulation is presented. It consists of the discretization of Laplace's problem by means of finite differences scheme of second order that gives a linear system solved by Preconditioned Conjugate Gradient Method (PCG). In this work, we observe that the performance and the accuracy of PCG solver depend on the grid resolution and the ratio of Laplace's coefficients. Moreover, a case study of an implementation of PCG solver with diagonal preconditioner on multi-core GPU architecture based on CUBLAS library is proposed. [ABSTRACT FROM AUTHOR]