Use Of Graphics Processing Units Research Articles

Characterizing and Exploiting Soft Error Vulnerability Phase Behavior in GPU Applications

System reliability has become a first-class design constraint. As the use of Graphics Processing Units (GPU) continues to increase in compute applications, including High Performance Computing (HPC) and safety-critical applications, so do the number of transient faults in GPUs. However, our understanding of the potential impact of transient fault propagation in GPU applications remains limited. This study shows that the resilience characteristics of GPU programs change significantly during program execution and these characteristics show repetitive, time-varying behavior. Interestingly, these repetitive, time-varying, resilience characteristics of GPU programs do not align or correlate well with the performance phases of GPU programs. Furthermore, this work discovers and validates that temporal changes in the vulnerability behavior during a kernel execution tends to coincide with changes in basic block execution paths. Finally, we demonstrate how these observations can be exploited to accelerate the fault injection campaigns for reliability assessment of GPU programs by an order of magnitude and open opportunities for designing other effective resilience mitigation strategies.

Toward large-scale simulation of residual stress and distortion in wire and arc additive manufacturing

This study aims to advance the structural analysis of wire and arc additive manufacturing (WAAM) by considering the thermomechanical features inherent in direct energy deposition. Simulation approaches including the iterative substructure method (ISM), dynamic mesh refining method (DMRM), and graphics processing unit (GPU) based explicit finite element method (FEM) were developed for accelerating additive manufacturing stress analysis that is very time consuming by conventional numerical methods. ISM and DMRM take advantage of the strong nonlinearity phenomenon near the moving heat source by reducing the global iterations and refining the local mesh, respectively. In addition to the use of GPUs, the explicit FEM is accelerated by a time scaling technique based on the inherent strain concept. The residual stress and distortion of two large builds were analyzed, showing very consistent numerical results and good agreement with experiments. Compared with the commercial software Abaqus, the novel approaches reduced the computational cost substantially without compromising accuracy. The GPU code showed the highest computational efficiency (30∼70×), while DMRM and ISM had acceleration factors of 9× and 3×, respectively. Such high-fidelity modeling approaches will be very useful for building up a digital twin of WAAM to reduce development time and cost.

GPU-Accelerated Automatic Microseismic Monitoring Algorithm (GAMMA) and Its Application to the 2019 Ridgecrest Earthquake Sequence

Abstract Foreshocks and/or aftershocks play critical roles in improving our understanding of the processes of faulting, such as nucleation of earthquakes, earthquake triggering, and postseismic deformation. A rapid and accurate earthquake detection and location algorithm can provide timely information of seismic activities, thereby benefitting our understanding of physical mechanisms of faulting and seismic hazard assessment. We have developed a graphic processing unit (GPU)-accelerated automatic microseismic monitoring algorithm (GAMMA) for accurate and near real-time detection and location of earthquakes. GAMMA utilizes methods based on backprojection to automatically detect potential earthquakes, and then the waveforms of qualified earthquakes are selected as templates when searching for small earthquakes in continuous recordings using the template-matching algorithm. The use of GPUs has substantially accelerated the calculations and has made GAMMA capable of (near-)real-time earthquake monitoring. We have successfully applied GAMMA to the 2019 Ridgecrest earthquake sequence in southern California. The number of earthquakes detected by GAMMA is more than 21 times that documented in the regional catalog. The more complete catalog determined by GAMMA may provide crucial information for improving our understanding of the physical mechanisms of faulting and also supply useful constraints for a variety of types of studies, including dynamic rupture simulations and crustal deformation modeling.

High-performance computing in water resources hydrodynamics

Abstract This work presents a vision of future water resources hydrodynamics codes that can fully utilize the strengths of modern high-performance computing (HPC). The advances to computing power, formerly driven by the improvement of central processing unit processors, now focus on parallel computing and, in particular, the use of graphics processing units (GPUs). However, this shift to a parallel framework requires refactoring the code to make efficient use of the data as well as changing even the nature of the algorithm that solves the system of equations. These concepts along with other features such as the precision for the computations, dry regions management, and input/output data are analyzed in this paper. A 2D multi-GPU flood code applied to a large-scale test case is used to corroborate our statements and ascertain the new challenges for the next-generation parallel water resources codes.

Efficient CT Image Reconstruction in a GPU Parallel Environment.

Computed tomography is nowadays an indispensable tool in medicine used to diagnose multiple diseases. In clinical and emergency room environments, the speed of acquisition and information processing are crucial. CUDA is a software architecture used to work with NVIDIA graphics processing units. In this paper a methodology to accelerate tomographic image reconstruction based on maximum likelihood expectation maximization iterative algorithm and combined with the use of graphics processing units programmed in CUDA framework is presented. Implementations developed here are used to reconstruct images with clinical use. Timewise, parallel versions showed improvement with respect to serial implementations. These differences reached, in some cases, 2 orders of magnitude in time while preserving image quality. The image quality and reconstruction times were not affected significantly by the addition of Poisson noise to projections. Furthermore, our implementations showed good performance when compared with reconstruction methods provided by commercial software. One of the goals of this work was to provide a fast, portable, simple, and cheap image reconstruction system, and our results support the statement that the goal was achieved.

A Doppler Correcting Software Defined Radio Receiver Design for Satellite Communications

We present a flexible software defined radio (SDR) receiver design for the reception and demodulation of satellite signals. The proposed SDR performs an online Doppler search, as well as a code-rate and code-phase search online. The use of graphics processing units (GPUs) allows the radio to compensate Doppler offsets and achieve symbol synchronization in real time. This eliminates the need for a priori information on the satellites position and velocity to perform Doppler compensation on the downlink. To successfully utilize the enhanced computational power of graphics processing units (GPUs) and achieve real-time performance, we utilize acausal batch based signal processing, which allows for the use of fast convolution algorithms. These can efficiently be parallalized and run on GPUs. The Doppler search and demodulation is done by using matched filters. This allows one to adapt the demodulator to work for many modulation schemes by merely changing the set of filters. The demodulator itself essentially works as a batched correlation receiver. Additionally, the proposed SDR can decode signals from multiple antennas that are located at geographically spread ground stations simultaneously. The decoded bits are compared using soft decisions. This can greatly increase the rejection of local burst interference as well as polarized interference and tracking signals that rotate in polarization.

A Parallel Finite-Element Time-Domain Method for Nonlinear Dispersive Media

In this article, a novel use of graphics processing units (GPUs) is presented for the acceleration of finite-element time-domain (FETD) methods containing electrically complex media. By leveraging the massively parallel architecture of the GPU via NVIDIA's Compute Unified Device Architecture (CUDA) language, the immense computational burden imposed by these materials can be largely alleviated, facilitating their modeling and incorporation into electromagnetic devices and systems. To that end, an analysis of both mixed and vector wave equation-based nonlinear dispersive FETD algorithms is presented in order to both identify computational bottlenecks and determine their amenability to parallelization. Based on this analysis, a parallel elemental matrix-evaluation procedure is proposed, which when coupled to the recently derived Gaussian belief propagation method for matrix assembly and solution, demonstrates a performance increase of up to 200 times as compared with a traditionally serial implementation.

GPU Accelerated Polar Harmonic Transforms for Feature Extraction in ITS Applications

Polar Harmonic Transform (PHT) is termed to represent a set of transforms those kernels are basic waves and harmonic in nature, which can improve the effect in Intelligent Transportation System (ITS) applications. PHTs consist of Polar Complex Exponential Transform (PCET), Polar Cosine Transform (PCT) and Polar Sine Transform (PST). PHTs can extract orthogonal and rotation invariant features and demonstrated superior performance in various image processing and computer vision applications. For real time systems and large multimedia databases, execution efficiency is always a significant challenge. With widespread use of Graphics Processing Unit (GPU), this study presents GPU based PHTs. Proposed methods are based on mathematical properties of PHTs and optimization techniques of GPU. Optimal parameter selections for GPU execution are also discussed. In our experiments, proposed methods are over 1800 times faster.

Nonadiabatic Molecular Dynamics on Graphics Processing Units: Performance and Application to Rotary Molecular Motors.

Nonadiabatic molecular dynamics (NAMD) simulations of molecular systems require the efficient evaluation of excited-state properties, such as energies, gradients, and nonadiabatic coupling vectors. Here, we investigate the use of graphics processing units (GPUs) in addition to central processing units (CPUs) to efficiently calculate these properties at the time-dependent density functional theory (TDDFT) level of theory. Our implementation in the FermiONs++ program package uses the J-engine and a preselective screening procedure for the calculation of Coulomb and exchange kernels, respectively. We observe good speed-ups for small and large molecular systems (comparable to those observed in ground-state calculations) and reduced (down to sublinear) scaling behavior with respect to the system size (depending on the spatial locality of the investigated excitation). As a first illustrative application, we present efficient NAMD simulations of a series of newly designed light-driven rotary molecular motors and compare their S1 lifetimes. Although all four rotors show different S1 excitation energies, their ability to rotate upon excitation is conserved, making the series an interesting starting point for rotary molecular motors with tunable excitation energies.

Real-time background oriented schlieren with self-illuminated speckle background

Background oriented schlieren (BOS) is a widely used technique that provides density gradient information in flow fields of interest. Raw BOS image data provide little, if any, visual indication of the density gradients in the flow. The acquired image data must be processed off-line after the test is completed, providing no indication of the quality of the BOS setup nor any feedback on the operational success of the test. The use of graphical processing units (GPUs) enables real-time acquisition, processing and display of BOS image data. In addition to real-time BOS processing, self-illuminated speckle backgrounds are implemented by incorporating a 4 K monitor into the BOS system. The combination of the real-time processing and a computer controlled/generated background speckle pattern enables self-optimization of the BOS system both prior and during testing.

Redesigning the rCUDA communication layer for a better adaptation to the underlying hardware

SummaryThe use of Graphics Processing Units (GPUs) has become a very popular way to accelerate the execution of many applications. However, GPUs are not exempt from side effects. For instance, GPUs are expensive devices which additionally consume a non‐negligible amount of energy even when they are not performing any computation. Furthermore, most applications present low GPU utilization. To address these concerns, the use of GPU virtualization has been proposed. In particular, remote GPU virtualization is a promising technology that allows applications to transparently leverage GPUs installed in any node of the cluster. In this paper, the remote GPU virtualization mechanism is comparatively analyzed across three different generations of GPUs. The first contribution of this study is an analysis about how the performance of the remote GPU virtualization technique is impacted by the underlying hardware. To that end, the Tesla K20, Tesla K40, and Tesla P100 GPUs along with FDR and EDR InfiniBand fabrics are used in the study. The analysis is performed in the context of the rCUDA middleware. It is clearly shown that the GPU virtualization middleware requires a comprehensive design of its communication layer, which should be perfectly adapted to every hardware generation in order to avoid a reduction in performance. This is precisely the second contribution of this work, ie, redesigning the rCUDA communication layer in order to improve the management of the underlying hardware. Results show that it is possible to improve bandwidth up to 29.43%, which translates into up to 4.81% average less execution time in the performance of the analyzed applications.

METHOD OF SOLVING SYSTEM OF LINEAR ALGEBRAIC EQUATIONS WITH USE OF HIGH PERFORMANCE COMPUTING ON GPU

The article discusses the use of graphics processing units for solving large system of linear algebraic equations (SLAE). A heterogeneous multiprocessor computing platform produced by the NIIVK, whose architecture allows the integration of general‑ purpose microprocessor modules with graphics processor modules was used as an equipment for solving SLAEs. The description of the SLAE solution program, developed on the basis of the CUBLAS CUDA software interface library, is given. A method is proposed for increasing the accuracy of calculations of linear systems based on the use of a modified Gauss method. It has been established that the use of the modified Gauss method practically does not increase the program operation time with a significant increase in the accuracy of calculations. It is concluded that the use of graphics processors for solving SLAEs allows processing matrices of a larger size compared to the use of general‑purpose microprocessors.

Parallelized Monte-Carlo dosimetry using graphics processing units to model cylindrical diffusers used in photodynamic therapy: From implementation to validation.

The Monte-Carlo method is the standard method for computing the dosimetry of both ionizing and non-ionizing radiation. Because this technique is highly time-consuming in conventional implementations, several improvements have recently been developed to speed-up simulations. Among the improvements, the use of graphics processing units (GPU) to parallelize algorithms provides a cost-efficient solution to accelerate the Monte-Carlo method. Parallel implementation of Monte-Carlo using GPU technology is described in the context of photodynamic therapy (PDT) dosimetry. This algorithm has been optimized to compute light emitted from optical fibers with cylindrical diffusers that are used in interstitial PDT applications. A comparison of the experimental measurements used to assess the results of the Monte-Carlo method is detailed. Illumination profiles of several commercially available diffusers are measured using an optical phantom that mimics the optical properties of the brain. Additionally, this Monte-Carlo method is compared to ex-vivo measurements made by a device dedicated to intraoperative PDT treatment of brain tumors. The results of the GPU Monte-Carlo validation are in accordance with the recommendations of the American Association of Physicists in Medicine. The acceleration obtained with the GPU implementation is in accordance with the literature and is sufficiently fast to be integrated in a treatment planning system dedicated to planning routine clinical interstitial PDT treatments.

GPU acceleration of splitting schemes applied to differential matrix equations

We consider differential Lyapunov and Riccati equations, and generalized versions thereof. Such equations arise in many different areas and are especially important within the field of optimal control. In order to approximate their solution, one may use several different kinds of numerical methods. Of these, splitting schemes are often a very competitive choice. In this article, we investigate the use of graphical processing units (GPUs) to parallelize such schemes and thereby further increase their effectiveness. According to our numerical experiments, large speed-ups are often observed for sufficiently large matrices. We also provide a comparison between different splitting strategies, demonstrating that splitting the equations into a moderate number of subproblems is generally optimal.

Combined Method of Visualization of Functionally Defined Surfaces and Three-Dimensional Textures

A combined method of object visualization based on analytical and scalar perturbation functions and three-dimensional textures with the use of graphics processing units is proposed. To display the terrain and the change in levels of detail, the same method as that for color textures is applied, and vertex shaders are used in the case of scattered light. A method of real-time visualization of volumetric clouds is described. For this purpose, it is proposed to form three-dimensional textures by means of pre-processing of the cloud structure and volume-oriented visualization.

Differential Evolution Implementation for Power Quality Disturbances Monitoring using OpenCL

This article presents a new methodology to implement a computational parallel scheme based on Differential Evolution (DE) algorithm through the use of Graphical Processing Units (GPU). ...

GPU-accelerated steady-state computation of large probabilistic Boolean networks

Abstract Computation of steady-state probabilities is an important aspect of analysing biological systems modelled as probabilistic Boolean networks (PBNs). For small PBNs, efficient numerical methods to compute steady-state probabilities of PBNs exist, based on the Markov chain state-transition matrix. However, for large PBNs, numerical methods suffer from the state-space explosion problem since the state-space size is exponential in the number of nodes in a PBN. In fact, the use of statistical methods and Monte Carlo methods remain the only feasible approach to address the problem for large PBNs. Such methods usually rely on long simulations of a PBN. Since slow simulation can impede the analysis, the efficiency of the simulation procedure becomes critical. Intuitively, parallelising the simulation process is the ideal way to accelerate the computation. Recent developments of general purpose graphics processing units (GPUs) provide possibilities to massively parallelise the simulation process. In this work, we propose a trajectory-level parallelisation framework to accelerate the computation of steady-state probabilities in large PBNs with the use of GPUs. To maximise the computation efficiency on a GPU, we develop a dynamical data arrangement mechanism for handling different size PBNs with a GPU. Specially, we propose a reorder-and-split method to handle both large and dense PBNs. Besides, we develop a specific way of storing predictor functions of a PBN and the state of the PBN in the GPU memory. Moreover, we introduce a strongly connected component (SCC)-based network reduction technique to further accelerate the computation speed. Experimental results show that our GPU-based parallelisation gains approximately a 600-fold speedup for a real-life PBN compared to the state-of-the-art sequential method.

Galaxy detection and identification using deep learning and data augmentation

Abstract We present a method for automatic detection and classification of galaxies which includes a novel data-augmentation procedure to make trained models more robust against the data taken from different instruments and contrast-stretching functions. This method is shown as part of AstroCV, a growing open source computer vision repository for processing and analyzing big astronomical datasets, including high performance Python and C++ algorithms used in the areas of image processing and computer vision. The underlying models were trained using convolutional neural networks and deep learning techniques, which provide better results than methods based on manual feature engineering and SVMs in most of the cases where training datasets are large. The detection and classification methods were trained end-to-end using public datasets such as the Sloan Digital Sky Survey (SDSS), the Galaxy Zoo, and private datasets such as the Next Generation Virgo (NGVS) and Fornax (NGFS) surveys. Training results are strongly bound to the conversion method from raw FITS data for each band into a 3-channel color image. Therefore, we propose data augmentation for the training using 5 conversion methods. This greatly improves the overall galaxy detection and classification for images produced from different instruments, bands and data reduction procedures. The detection and classification methods were trained using the deep learning framework DARKNET and the real-time object detection system YOLO. These methods are implemented in C language and CUDA platform, and makes intensive use of graphical processing units (GPU). Using a single high-end Nvidia GPU card, it can process a SDSS image in 50 ms and a DECam image in less than 3 s. We provide the open source code, documentation, pre-trained networks, python tutorials, and how to train your own datasets, which can be found in the AstroCV repository. https://github.com/astroCV/astroCV .

GPU-based processing of Hartmann–Shack images for accurate and high-speed ocular wavefront sensing

Hartmann–Shack aberrometry is a widely used technique in the field of visual optics but, high-speed and accurate processing of Hartmann–Shack images can be a computationally expensive/resource intensive task. While some advancements have been made in achieving high-performance processing units, they have not been specifically designed for processing Hartmann–Shack images of the human eye with Graphics Processing Units. In this work, we present the first full-Graphics Processing Unit implementation of a Hartmann–Shacksensor algorithm aimed at accurately measuring ocular aberrations at a high speed from high-resolution spot pattern images. The proposed algorithm, called PaPyCS (Parallel Pyramidal Centroid Search), is inherently parallel and performs a very robust centroid search to avoid image noise and other artifacts. This is a field where the use of Graphics Processing Units have not been exploited despite the fact that they can boost Adaptive Optics systems and related closed-loop approaches. Our proposed implementation achieves processing speeds of 380 frames per second for high resolution (1280x1280 pixels) images, in addition to showing a high resilience to system and image artifacts that appear in Hartmann–Shack images from human eyes: more than 98% of the Hartmann–Shack images, with aberrations of up to 4 μ m Root Mean Square for a 5.12mm pupil diameter, were measured with less than 0.05 μ m Root Mean Square Error, which is basically negligible for ocular aberrations.

Optimizing photoacoustic image reconstruction using cross-platform parallel computation

Three-dimensional (3D) image reconstruction involves the computations of an extensive amount of data that leads to tremendous processing time. Therefore, optimization is crucially needed to improve the performance and efficiency. With the widespread use of graphics processing units (GPU), parallel computing is transforming this arduous reconstruction process for numerous imaging modalities, and photoacoustic computed tomography (PACT) is not an exception. Existing works have investigated GPU-based optimization on photoacoustic microscopy (PAM) and PACT reconstruction using compute unified device architecture (CUDA) on either C++ or MATLAB only. However, our study is the first that uses cross-platform GPU computation. It maintains the simplicity of MATLAB, while improves the speed through CUDA/C++ − based MATLAB converted functions called MEXCUDA. Compared to a purely MATLAB with GPU approach, our cross-platform method improves the speed five times. Because MATLAB is widely used in PAM and PACT, this study will open up new avenues for photoacoustic image reconstruction and relevant real-time imaging applications.