Compute Unified Device Architecture Technology Research Articles

Development of software and algorithms of parallel learning of artificial neural networks using CUDA technologies

The object of research is to parallelize the learning process of artificial neural networks to automate the procedure of medical image analysis using the Python programming language, PyTorch framework and Compute Unified Device Architecture (CUDA) technology. The operation of this framework is based on the Define-by-Run model. The analysis of the available cloud technologies for realization of the task and the analysis of algorithms of learning of artificial neural networks is carried out. A modified U-Net architecture from the MedicalTorch library was used. The purpose of its application was the need for a network that can effectively learn with small data sets, as in the field of medicine one of the most problematic places is the availability of large datasets, due to the requirements for data confidentiality of this nature. The resulting information system is able to implement the tasks set before it, contains the most user-friendly interface and all the necessary tools to simplify and automate the process of visualization and analysis of data. The efficiency of neural network learning with the help of the central processor (CPU) and with the help of the graphic processor (GPU) with the use of CUDA technologies is compared. Cloud technology was used in the study. Google Colab and Microsoft Azure were considered among cloud services. Colab was first used to build a prototype. Therefore, the Azure service was used to effectively teach the finished architecture of the artificial neural network. Measurements were performed using cloud technologies in both services. The Adam optimizer was used to learn the model. CPU duration measurements were also measured to assess the acceleration of CUDA technology. An estimate of the acceleration obtained through the use of GPU computing and cloud technologies was implemented. CPU duration measurements were also measured to assess the acceleration of CUDA technology. The model developed during the research showed satisfactory results according to the metrics of Jaccard and Dyce in solving the problem. A key factor in the success of this study was cloud computing services.

Massively parallel solution of the Lambert problem

This paper considers the possibility of massively parallel solution to the Lambert problem on graphics processing units. Several of the most popular solution algorithms were used for this problem. Software implementation of algorithms under consideration has been developed using Compute Unified Device Architecture technology that allows performing calculations on graphics processing units. A distinctive feature of this project is an attempt to use parallel programming within an iterative scheme. The developed algorithms have been tested on a number of typical tasks: contouring of Earth – Mars direct flight isolines and solution of the problem of a flight to a group of asteroids for the given start date and flight duration. For such tasks we give estimates of execution time and paralleling efficiency.

NUMERICAL IMPLEMENTATION OF A PARALLEL ALGORITHM FOR SOLVING THE PROBLEM OF POLLUTANT TRANSPORT IN A RESERVOIR ON A HIGH-PERFORMANCE COMPUTER SYSTEM

The purpose of this work is to create a software package for a distributed solution of the problem of transporting a pollutant in a reservoir with complex bathymetry and the presence of technological structures. An algorithm has been developed for the parallel solution of the problem of transporting a pollutant (pollutant) in a reservoir on a graphics accelerator controlled by the CUDA (Compute Unified Device Architecture) system; a comparative analysis of the operation of algorithms on a CPU (Central Processing Unit) and on a graphics accelerator GPU (Graphics Processing Unit) made it possible to evaluate their performance. The software implementation of the modules included in the complex is described, the main classes and implemented methods are documented. The results of numerical experiments showed that solving of pollutant transport’s problem based on the CUDA technology is ineffective for small grids (up to 100 ´ 100 computational nodes). In the case of large grids (1000 ´ 1000 computational nodes), the use of CUDA technology reduces the computation time by an order of magnitude. An analysis of the experiments carried out with the developed components of software showed that the maximum value of the ratio of the algorithm operating time that implements the set task of transferring matter in a shallow water on a GPU to the operating time of a similar algorithm on the CPU was 24.92 times, which is achieved on a grid of 1000 ´ 1000 computational nodes. Implementation of methods for decomposition of grid regions is proposed for solving computationally laborious problems of diffusion-convection, including the problem of transporting pollutants in a reservoir with complex bathymetry with technological objects that take into account the architecture and parameters of a MSC (Multiprocessor Computing System) located on the basis of the infrastructure facility of the STU (Scientific and Technological University) “Sirius” (Sochi, Russia). Consideration was made for such a property of a computing system as the time it takes to transmit and receive floating point data. An algorithm for the parallel solution of the task under the control of MPI (Message Passing Interface) technology has been developed, and its efficiency has been assessed. The acceleration values of the proposed algorithm are obtained depending on the number of involved computers (processors) and the size of the computational grid. The maximum number of computers used is 24, the maximum size of the computational grid was 10 000 ´ 10 000 computational nodes. The developed algorithm showed low efficiency for small computational grids (up to 100 ´ 100 computational nodes). In the case of large computational grids ( from 1000  1000 computational nodes), the use of MPI reduces the computation time by several times.

A hybrid MPI-CUDA approach for nonequispaced discrete Fourier transformation

Nonequispaced discrete Fourier transformation (NDFT) is widely applied in all aspects of computational science and engineering. The computational efficiency and accuracy of NDFT has always been a critical issue in hindering its comprehensive applications both in intensive and in extensive aspects of scientific computing. In our previous work Yang et al. (2018), a CUNFFT method was proposed and it shown outstanding performance in handling NDFT at intermediate scale based on CUDA (Compute Unified Device Architecture) technology. In the current work, we further improved the computational efficiency of the CUNTTF method using an efficient MPI-CUDA hybrid parallelization (HP) scheme of NFFT to achieve a cutting-edge treatment of NDFT at super extended scale. Within this HP-NFFT method, the spatial domain of NDFT is decomposed into several parts according to the accumulative feature of NDFT and the detailed number of CPU and GPU nodes. These decomposed NDFT subcells are independently calculated on different CPU nodes using a MPI process-level parallelization mode, and on different GPU nodes using a CUDA thread-level parallelization mode and CUNFFT algorithm. A massive benchmarking of the HP-NFFT method indicates that this method exhibits a dramatic improvement in computational efficiency for handling NDFT at super extended scale without loss of computational precision. Furthermore, the HP-NFFT method is validated via the calculation of Madelung constant of fluorite crystal structure, and thereafter verified that this method is robust for the calculation of electrostatic interactions between charged ions in molecular dynamics simulation systems. Program summaryProgram title: HP-NFFTCPC Library link to program files:http://dx.doi.org/10.17632/ys2y92jkwy.1Licensing provisions: GNU General Public License 3Programming language: MPI, C, and CUDA CSupplementary material: The program is designed for effective computation of large-scale nonequispaced discrete Fourier transformation (NDFT), which runs on particular computers equipped with NVIDIA GPUs. It has been tested on (a) one single computer node with Intel(R) Core(TM) i7-3770 @ 3.40 GHz (CPU) and GTX 980 Ti (GPU), and (b) MPI parallel computer nodes with the same configurations.Nature of problem: For NDFT, the computation is extremely time-consuming in many domains of computational physics due to the failure to utilize FFT directly, which often affects the computational efficiency of whole system seriously. Although the parallel method CUNFFT based on GPU has achieved a qualitative leap compared with previous methods in NDFT computation, the computation capability is limited to the throughput capacity of GPU when the size of NDFT system is big enough.Solution method: We constructed a hybrid parallel architecture, in which CPU and GPU are combined to accelerate the NDFT computation effectively. Firstly, the NDFT system is divided into several subcells via domain-decomposition method. Then MPI (Message Passing Interface) is used to implement the CPU-parallel computation with each computer node corresponding to a particular subcell, and furthermore each subcell in one computer node will be executed in GPU in parallel efficiently. In this hybrid parallel method, the most critical technical problem is how to parallelize a CUNFFT in the parallel strategy, which is conquered effectively by deep-seated research of basic principles and some algorithm skills.Restrictions: The HP-NFFT is mainly oriented to, and has shown significant computational efficiency to deal with large-scale NDFT computations. However, for a small NDFT system containing less than 10^6 particles, the mode of multiple computer nodes has no apparent efficiency advantage or even lower efficiency compared with the mode of a single computer node because of the serious network delay.

Strength Check of Aircraft Parts Based on Multi-GPU Clusters for Fast Calculation of Sparse Linear Equations

In order to improve the cost-effectiveness ratio, the next-generation vehicle needs to meet the requirements of reuse, while adopting a lighter structural weight, so it is necessary to realize the strength calculation and condition monitoring of key components in the digital twin. Most of the current monitoring methods are based on the characteristics of various data acquisition systems, but they need the support of a large number of flight data. The disadvantages of the above strategy can be avoided by reducing the structure of aircraft components to a finite element model and quickly checking the key components in the health management system. In order to solve the problem of fast calculation of the finite element model of the key components of the aircraft, a parallel algorithm and framework of large-scale sparse matrix preprocessing conjugate gradient method based on CUDA(Compute Unified Device Architecture) technology is proposed in the multi GPU(Graphics Processing Unit) workstation cluster environment. Once the sparse matrix is too large to be processed in a single workstation, this paper discusses how to realize the optimized data segmentation in the distributed multi-GPU computing environment. For the problem of iterative solution of matrix preprocessing, two preprocessing strategies of matrix bandwidth reduction parallelization and incomplete Cholesky decomposition are proposed, and asynchronous task concurrency and load balancing strategies are designed on the architecture. The calculation of some examples in the standard sparse matrix database shows that the algorithm and architecture proposed in this paper have the ability to solve large-scale sparse matrix quickly and efficiently, and can complete the fast strength verification of vehicle components.

High performance filtering for big datasets from Airborne Laser Scanning with CUDA technology

There are many studies on the problems of processing big datasets provided by Airborne Laser Scanning (ALS). The processing of point clouds is often executed in stages or on the fragments of the measurement set. Therefore, solutions that enable the processing of the entire cloud at the same time in a simple, fast, efficient way are the subject of many researches. In this paper, authors propose to use General-Purpose computation on Graphics Processing Units (GPGPUs) to process the big datasets obtained from ALS. GPGPU handles computation for computer graphics using GPUs (Graphic Processing Units). This study was based on programming model Compute Unified Device Architecture (CUDA), which facilitates the development of applications in GPUs. CUDA programming was used to carry out the filtration based on adaptive TIN model method in the initial stage of the processing of big ALS dataset. Results of the analysis showed that GPGPU can be used for the filtration of ALS point clouds and significantly speeds up calculations for big dataset.

A new theoretical derivation of NFFT and its implementation on GPU

Abstract An efficient calculation of NFFT (nonequispaced fast Fourier transforms) is always a challenging task in a variety of application areas, from medical imaging to radio astronomy to chemical simulation. In this article, a new theoretical derivation is proposed for NFFT based on gridding algorithm and new strategies are proposed for the implementation of both forward NFFT and its inverse on both CPU and GPU. The GPU-based version, namely CUNFFT, adopts CUDA (Compute Unified Device Architecture) technology, which supports a fine-grained parallel computing scheme. The approximation errors introduced in the algorithm are discussed with respect to different window functions. Finally, benchmark calculations are executed to illustrate the accuracy and performance of NFFT and CUNFFT. The results show that CUNFFT is not only with high accuracy, but also substantially faster than conventional NFFT on CPU.

Parallelized TSM-RT Method for the Fast RCS Prediction of the 3-D Large-Scale Sea Surface by CUDA

In this paper, a parallelized two-scale model-ray tracing (TSM-RT) method is developed for the fast radar cross-section prediction of the 3-D large-scale sea surface by the compute unified device architecture (CUDA). The TSM-RT method is an accelerated RT method, where the sea surface is constructed into a great deal of large triangles, which consist of many small triangles; this is called the two-scale model. For the TSM-RT method, the large triangles are utilized to determine the asymptotical ray paths, while the small triangles are utilized to calculate the scattered far fields. In addition, the CUDA of NVIDIA takes advantage of the graphics processing units (GPUs) for parallel computing, and greatly improves the speed of computation. The scattering field of each ray in the TSM-RT method is calculated independently. Therefore, a parallelization concept, which is based on the CUDA technology, is presented to improve the computational efficiency. The asynchronous transfer and shared memory are used to further improve the speedup factors. Our method is validated by comparing the numerical results with those obtained using a CPU. At the same time, significant speedup factors are achieved compared with the CPU-based TSM-RT.

Simulation of digital filter banks and signal classification in wideband monitoring tasks

The paper is devoted to handling radio monitoring tasks (wideband monitoring tasks) in real time. We suggest a new multichannel modification of the weighted overlap-add algorithm (WOLA-algorithm) for processing vector (multichannel) signals and performing multichannel signal classification. Filter bank implementation is considered using computers based on the CUDA (Compute Unified Device Architecture) technology. We show that CUDA is efficient for large signal sets due to its low temporal costs. We study signal classification in filter bank channels for different signal-tonoise ratios using binary decision trees, the iterative procedure Adaboost, and neural networks. Our experiments provided a total classification error less than 10%.

Acceleration of JPEG Decoding Process using CUDA

In this paper we have implemented efficient JPEG (Joint Photographic experts group) Decoder on GPU (Graphic Processing Unit) using NVIDIA CUDA (Compute Unified Device Architecture) Technology. This decoder is capable of decoding images of Ultra HD resolution with superfast speed GPU is used to assist the CPU for time consuming tasks. In this paper IDCT module which consumes 70 to 80 percent of computation time is implemented on GPU. An asynchronous parallel execution between the CPU and the GPU is used at a same time to improve the JPEG decoder acceleration rate. In this work, the JPEG decoder based on the CUDA performs decompression of images of size 2560 x 1600 pixels and below. Finally the results are shown with respect to different sized images and consumed time for decoding. The results show that this decoder faster in multiple times than the decoder in CPU. General Terms:

Optimized parallel implementation of face detection based on GPU component

Face detection is an important aspect for various domains such as: biometrics, video surveillance and human computer interaction. Generally a generic face processing system includes a face detection, or recognition step, as well as tracking and rendering phase. In this paper, we develop a real-time and robust face detection implementation based on GPU component. Face detection is performed by adapting the Viola and Jones algorithm. We have developed and designed optimized several parallel implementations of these algorithms based on graphics processors GPU using CUDA (Compute Unified Device Architecture) description.First, we implemented the Viola and Jones algorithm in the basic CPU version. The basic application is widened to GPU version using CUDA technology, and freeing CPU to perform other tasks. Then, the face detection algorithm has been optimized for the GPU using a grid topology and shared memory. These programs are compared and the results are presented. Finally, to improve the quality of face detection a second proposition was performed by the implementation of WaldBoost algorithm.

Graphics processor unit accelerated finite-difference time domain method for electromagnetic scattering from one-dimensional large scale rough soil surface at low grazing incidence

The graphics processor unit-based finite-difference time domain (FDTD) algorithm is applied to study the electromagnetic (EM) scattering from one-dimensional (1-D) large scale rough soil surface at a low grazing incident angle. The FDTD lattices are truncated by a uniaxial perfectly matched layer, and finite difference equations are employed in the whole computation domain for convenient parallelization. Using Compute Unified Device Architecture technology, we achieve significant speedup factors. Also, shared memory and asynchronous transfer are used to further improve the speedup factors. Our method is validated by comparing the numerical results with those obtained by using a CPU. The influences of the incident angle, correlation length l, and root-mean-square height δ on the bistatic scattering coefficient of a 1-D large scale rough surface at low grazing incidence are also discussed.

Low-coherence interferometry with polynomial interpolation on Compute Unified Device Architecture-enabled graphics processing units

An algorithm for interpolation of central fringe position in low-coherence interferometry measurements is presented. The algorithm is based on a polynomial curve fitting. Fast calculation of interpolation is possible due to the use of an NVIDIA Compute Unified Device Architecture (CUDA) technology, which allows independent analysis of different points of a high-resolution detector matrix on separate cores of a graphics processing unit (GPU). The dependency of the method’s accuracy on the spectral width of the light source is checked. The computation times on a GPU are compared with those achieved with a multicore central processing unit, showing nearly 30 times faster calculations when using CUDA technology. The algorithm accuracy is tested by measuring a flat glass surface with two different cameras—an ordinary CCD camera and a cooled EMCCD camera. Finally, the algorithm is applied to measurements of a populated optical fiber connector array prototyped using deep proton writing technology.

JPEG2000 Image Compression Method Based on GPGPU

In order to improve the compression speed of JPEG2000, the JPEG2000 compression standard is analysised and it concluded that the part data of the core algorithm that is DWT in JPEG2000 are independent from each other, so it is very suitable for parallel processing. CUDA (Compute Unified Device Architecture) is a latest software and hardware exploitation platform released by NVIDIA which is very suitable for large-scale data parallel computing.Using CUDA technology on general purpose graphic process unit (GPGPU) could speed up DWT algorithm parallelly and the program is optimized based on the characteristics of GPGPU storage space. The obtained experimental results show the DWT algorithm that is optimized by CUDA parallelly can improve the computing speed.

Large-scale CFD–DEM simulations of fluidized granular systems

The combination of Computational Fluid Dynamics (CFD) and Discrete Element Model (DEM) is a powerful tool for studying fluidized particulate systems and granular flows. In DEM, the individual interaction forces between particles are treated on a particle–particle pair basis, and therefore, this method is computational expensive. In addition, the CFD-calculation of the fluid flow increases the computational effort. Thus, current CFD–DEM simulations are limited to systems with particle numbers not exceeding 105. In order to simulate realistic systems, the recently available Compute Unified Device Architecture (CUDA) technology can be applied, which can perform massively-parallel DEM-simulations with several million particles on a single desk-side Graphics Processing Unit (GPU). The objective of this work is to present a new hybrid approach to solve CFD–DEM problems in gas–solid fluidized beds systems applying an efficient coupling method suitable for large-scale simulations. We are using the CUDA technology for the particle simulation and introducing a coupling methodology with a commercial CFD-code. The coupling method between a CFD-code, running on the CPU and our CUDA-based DEM-code running on the GPU, is introduced and discussed. The numerical results are compared to the CFD–DEM and the experimental results of Van Buijtenen et al. (2011). A good agreement was achieved. Finally, fluidized system simulations with up to 25 million particles are presented, which is an unprecented number.

An application of graphical numerical accelerators in simulations of ion-transport through biological membranes

The modeling of ion-transport through biological membranes is important for understanding many life processes. The transmembrane potential and ion concentrations in the stationary state can be measured in in-vivo experiments. They can also be simulated within membrane models. Here we consider a basic model of ion transport that describes the time evolution of ion concentrations and potentials through a set of nonlinear ordinary differential equations. To reduce the computation time I have developed an application for simulation of the ion-flows through a membrane starting from an ensemble of initial conditions, optimized for a Graphical Processing Unit (GPU). The application has been designed for the CUDA (Compute Unified Device Architecture) technology. It is written in CUDA C programming language and runs on NVIDIA TESLA family of numerical accelerators. The calculation speed can be increased almost 1000 times compared with a sequential program running on the Central Processing Unit (CPU) of a typical PC.

A Novel Method of Parallel Computation for the Whole Scene Test in Power System

In this paper, design and implementation of a new parallel computing method based on CUDA (Compute Unified Device Architecture) platform is described in detail. The method includes algorithm of matrix fraction and partial LU decomposition that are used to support parallel computing for simulation of the whole scene test in power system. The paper describes all the steps of algorithm implementation for deploying the software with CUDA technology. And performances are evaluated by test, in which two Jacobi matrix equations of power flow calculation with different dimension are selected as the test case. The results are compared with that of serial computing and MPI (Message Passing Interface). The results show that the method proposed in this paper has better performance of acceleration on CUDA platform, which can support the simulation of the whole scene test on wide-area power grid.

Real-Time Artifact-Free Image Upscaling

The problem of creating artifact-free upscaled images appearing sharp and natural to the human observer is probably more interesting and less trivial than it may appear. The solution to the problem, often referred to also as "single-image super-resolution," is related both to the statistical relationship between low-resolution and high-resolution image sampling and to the human perception of image quality. In many practical applications, simple linear or cubic interpolation algorithms are applied for this task, but the results obtained are not really satisfactory, being affected by relevant artifacts like blurring and jaggies. Several methods have been proposed to obtain better results, involving simple heuristics, edge modeling, or statistical learning. The most powerful ones, however, present a high computational complexity and are not suitable for real-time applications, while fast methods, even if edge adaptive, are not able to provide artifacts-free images. In this paper, we describe a new upscaling method (iterative curvature-based interpolation) based on a two-step grid filling and an iterative correction of the interpolated pixels obtained by minimizing an objective function depending on the second-order directional derivatives of the image intensity. We show that the constraints used to derive the function are related with those applied in another well-known interpolation method, providing good results but computationally heavy (i.e., new edge-directed interpolation (NEDI). The high quality of the images enlarged with the new method is demonstrated with objective and subjective tests, while the computation time is reduced of one to two orders of magnitude with respect to NEDI so that we were able, using a graphics processing unit implementation based on the nVidia Compute Unified Device Architecture technology, to obtain real-time performances.