Compute Unified Device Architecture NVIDIA CUDA

Sparse coding of image patches is a compact but computationally expensive method of representing images. As part of our SenSIP consortium industry projects, we implement the Orthogonal Matching Pursuit algorithm using a single CUDA kernel on a GPU and sparse codes for image patches are obtained in parallel. Image-based "exact search" and "visually similar search" using the image patch sparse codes are performed. Results demonstrate large speed-up over CPU implementations and achieve good retrieval performance.

Spiking neural network (SNN) models describe key aspects of neural function in a computationally efficient manner and have been used to construct large-scale brain models. Large-scale SNNs are challenging to implement, as they demand high-bandwidth communication, a large amount of memory, and are computationally intensive. Additionally, tuning parameters of these models becomes more difficult and time-consuming with the addition of biologically accurate descriptions. To meet these challenges, we have developed CARLsim 3, a user-friendly, GPU-accelerated SNN library written in C/C++ that is capable of simulating biologically detailed neural models. The present release of CARLsim provides a number of improvements over our prior SNN library to allow the user to easily analyze simulation data, explore synaptic plasticity rules, and automate parameter tuning. In the present paper, we provide examples and performance benchmarks highlighting the library's features.

Abstract—CUDA is a platform developed by Nvidia for general purpose computing on Graphic Processing Unit to utilize the parallelism capabilities. Serpent encryption is considered to have high security margin as its advantage; however it lacks in speed as its disadvantage. We ...

The present paper discusses digital filter banks in the context of wideband radio monitoring tasks including DFT-modulated filter banks and the weighted overlap-add (WOLA) algorithm. Filter bank software-hardware implementations are considered on the basis of Central Processing Unit (CPU) and Compute Unified Device Architecture (CUDA) with the use of Graphics Processing Unit (GPU). The main filter bank forms are studied and compared by their execution time in MATLAB. It is shown that CUDA technology is efficient for processing large datasets and outperforms computation results on CPU.

To the existence and influence, health related parameters and issues are at most importance to man. Various systems have been developed that are able to capture and monitor changes in health parameters. A real time remote monitoring of heart rate is presented in this paper. This system uses an alert and LCD display that are capable of monitoring the heart rate. A low cost, efficient and flexible heart rate detection and alert system using wireless module has been implemented in this paper. The sensors sense and measure the heart rate and detected signals are sent to control unit for further processing. The controller displays the heart rate on LCD which is then preceded to alert system. If there is a large difference between the normal and measured heart rates, then an alert will be provided by the system. This system is continuous, real time, safe and accurate in monitoring the heart rates. In this system the further application is attached for preventing system that name was AED (Automatic External Defibrillator). In case any low level beat is detects the defibrillator will give the shock on external body of the patients. After the shock the notification is send to the corresponding doctor. And voices also speak at the ICU unit if the patient is admitted in hospital.

As both CPU and GPU become employed in a wide range of applications, it has been acknowledged that both of these processing units (PUs) have their unique features and strengths and hence, CPU-GPU collaboration is inevitable to achieve high-performance computing. This has motivated significant amount of research on heterogeneous computing techniques, along with the design of CPU-GPU fused chips and petascale heterogeneous supercomputers. In this paper, we survey heterogeneous computing techniques (HCTs) such as workload-partitioning which enable utilizing both CPU and GPU to improve performance and/or energy efficiency. We review heterogeneous computing approaches at runtime, algorithm, programming, compiler and application level. Further, we review both discrete and fused CPU-GPU systems; and discuss benchmark suites designed for evaluating heterogeneous computing systems (HCSs). We believe that this paper will provide insights into working and scope of applications of HCTs to researchers and motivate them to further harness the computational powers of CPUs and GPUs to achieve the goal of exascale performance.

A B S T R A C T With the technology development of medical industry, processing data is expanding rapidly and computation time also increases due to many factors like 3D, 4D treatment planning, the increasing sophistication of MRI pulse sequences and the growing complexity of algorithms. Graphics processing unit (GPU) addresses these problems and gives the solutions for using their features such as, high computation throughput, high memory bandwidth, support for floating-point arithmetic and low cost. Compute unified device architecture (CUDA) is a popular GPU programming model introduced by NVIDIA for parallel computing. This review paper briefly discusses the need of GPU CUDA computing in the medical image analysis. The GPU performances of existing algorithms are analyzed and the computational gain is discussed. A few open issues, hardware configurations and optimization principles of existing methods are discussed. This survey concludes the few optimization techniques with the medical imaging algorithms on GPU. Finally, limitation and future scope of GPU programming are discussed. 1. Introduction Computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET) and ultrasound are famous medical modalities that produce the 2D, 3D and 4D types of medical images which are guiding the diagnosis process and treatment planning. The medical image processing and analysis are computationally expensive while medical imaging data dimension increasing [1]. The conventional CPU with limited multi-core is not sufficient to process these types of huge data. Graphics processing unit (GPU) is a new technology capable for finding out solutions to the computational problems in all the engineering and medical fields. In the medical industry, GPU is more suitable for processing the higher dimension data. GPU computation has provided a huge edge over the central processing units (CPU) with respect to computation speed. GPU is highly parallel, multithread, multiple core processors and has high memory bandwidth to give the solution to the computational problems [2]. The main reason for the evolution of powerful GPUs is the constant demand for greater realism in computer games. During the past few decades, the computational performance of GPUs has increased much more quickly than that of conventional CPUs. Hence it plays a major role in the field of modern industrial research and development. GPU has already achieved a significant speed (2x-1000x) than CPU implementation on various fields [3] [4] [5]. GPU is well suitable to implement the program execution with the different data elements. This process is called as data parallelism. Data parallelism is maps data elements to parallel threads available in GPU [6]. Data parallelism gives high gains in independent processes between data elements. The prime areas of data parallelism are 3D rendering, stereo vision, pattern recognition, image, video and medical industry applications. A large performance gap occurs between GPU and general purpose multi-core CPU. Architectural level comparison of CPU and GPU are given in Fig. 1. The design of a CPU is optimized for sequential programming. It makes use of sophisticated control logic to allow instructions from a single thread of execution to execute in parallel or even out of their sequential order while maintaining the appearance of sequential execution. Modern CPU microprocessors typically have four large processor cores designed to deliver strong sequential code performance but not enough to process the huge data. A basic model of GPU has large number of processor cores, ALU's, control units and various types of memories. In general, heterogeneous CPU and GPU computation is appreciable instead of standalone CPU or GPU implementation. The dependent processes are recommended in CPU and the independent processes can be accelerated by the GPU. GPU with high amount of threads give better performance. This paper reviews the implication of GPU programming model in medical image analysis and illustrated some applications with examples. The general framework of medical image analysis pipeline is given in Fig. 2. The computational complexities of all these fields are increasing

In this paper we have analyzed the comparison of radix sort algorithm on sequential and parallel procedures across three programming language platforms namely C, OpenMP based C++ and CUDA programming fixtures. The importance of radixsort to be a non comparison based sorting algorithm has been truncated. The algorithmic flow of data from the unsorted input to the desired output has also been shown using flowchart. Next to it, the theoretical complexity for radix sort is being illustrated and experimented for varied datasets of non-fixed sizes with varied bucket size ranges. Towards the end, the algorithmic complexities for radix sort have also been evaluated for the three algorithmic procedures used. The comparative efficiencies of radix sort for sorting a list of integers of different sizes and ranges have been evaluated. On the basis of analysis we have concluded the CUDA program to be approximately 100 times faster than the sequential C program and the multi-threaded parallel program enabled by OpenMP in C++ to be approximately 10 times faster than the sequential C program to sort a varied sized text file according to radix sort in an increasing order. This paper is a review of the serial and parallel algorithm implementation for sorting anonymously large dataset which can be used for graph explorations implemented over network analysis, which is also the future proposed work for our analytic study. Categories and Subject Descriptors F.2.2 [ANALYSIS OF ALGORITHMS AND PROBLEM COMPLEXITY]: Nonnumerical Algorithms and Problems –  Complexity of proof procedures, Sorting and searching.

The present paper discusses radio monitoring tasks and their solution using DFT-modulated filter banks. Filter bank software-hardware implementations are studied on the basis of Central Processing Unit (CPU) and Compute Unified Device Architecture (CUDA) with the use of Graphics Processing Unit (GPU). It is shown that CUDA technology is efficient for processing large datasets and outperforms computational results on CPU. The paper also considers signal classification in real time for different signal-to-noise ratios using a binary tree together with the iterative AdaBoost technique. Experiments show that it is possible to reach the total classification error of 10% for signals handled in radio monitoring tasks.

Cell metabolism is the biochemical machinery that provides energy and building blocks to sustain life. Understanding its fine regulation is of pivotal relevance in several fields, from metabolic engineering applications to the treatment of metabolic disorders and cancer. Sophisticated computational approaches are needed to unravel the complexity of metabolism. To this aim, a plethora of methods have been developed, yet it is generally hard to identify which computational strategy is most suited for the investigation of a specific aspect of metabolism. This review provides an up-to-date description of the computational methods available for the analysis of metabolic pathways, discussing their main advantages and drawbacks. In particular, attention is devoted to the identification of the appropriate scale and level of accuracy in the reconstruction of metabolic networks, and to the inference of model structure and parameters, especially when dealing with a shortage of experimental measurements. The choice of the proper computational methods to derive in silico data is then addressed, including topological analyses, constraint-based modeling and simulation of the system dynamics. A description of some computational approaches to gain new biological knowledge or to formulate hypotheses is finally provided.

This paper presents a Graphics Processing Unit (GPU) based parallel implementation for the All Pairs Shortest Paths problem. The implementation is based on the Floyd-Warshall algorithm and takes full advantage of the highly multithreaded architecture of current manycore GPUs. Our solution reduces the communication between CPU and GPU, improves the Streaming Multiprocessors (SMs) utilization, and makes intensive use of coalesced memory access to optimize graph data access. The advantage of the proposed implementation is demonstrated for several graphs randomly generated using the widely known graph library GTgraph. Graphs containing thousands of vertices and different edges densities, varying from sparse to complete graphs, were generated and used in the experiments. The results showed excellent performance on diverse graphs, achieving gains of up to 149x when compared to a sequential implementation, and outperforming state-of-the-art GPU implementations by a factor of almost four times. Our results confirm that GPU implementations can be competitive even for graph algorithms whose memory accesses and work distribution are both irregular and data-dependent.

The paper is devoted to the development of a WOLA-algorithm (weighted overlap add algorithm) for processing vector (multichannel) signals. The algorithm is considered as a generalization of one-dimensional WOLA with certain modifications. WOLA is expounded as an algorithm for real-time signal processing and the main advantages of WOLA are provided. We also discuss software-hardware implementation of WOLA using CPU (Central Processing Unit) and CUDA (Compute Unified Device Architecture). Finally we demonstrate the possibility of reducing hardware costs for multichannel signal processing using FPGA (field programmable gate arrays) and distributed arithmetic.

We present a fast and accurate 3D hand tracking method which relies on RGB-D data. The method follows a model based approach using a hierarchical particle filter variant to track the model’s state. The filter estimates the probability density function of the state’s posterior. As such, it has increased robustness to observation noise and compares favourably to existing methods that can be trapped in local minima resulting in track loses. The data likelihood term is calculated by measuring the discrepancy between the rendered 3D model and the observations. Extensive experiments with real and simulated data show that hand tracking is achieved at a frame rate of 90fps with less that 10mm average error using a GPU implementation, thus comparing favourably to the state of the art in terms of both speed and tracking accuracy.

In recent years the graphic processing units (GPUs) programmability has increased and this lead to use in several areas. GPUs can tackle enormous data parallel issues at a higher speed than the conventional CPU. Moreover, GPUs considered more affordable and energy-efficient than distributed systems. This paper gives a comprehensive review of the several published studies in the area of the parallel-shared memory based on the GPUs in the last few years. In addition, we represent the necessary computation time and speedup gains provided by different parallel-shared memory strategies. Moreover, in this article, a clear view and a detailed summary of such used algorithms/methods, hardware, and the results obtained of various parallel-shared memory using GPU approaches present in the literature. The maximum speedup as a performance is the main target of such researches where correlation algorithm and techniques were developed in experimental studies are reported.

GPGPUs have recently emerged as powerful vehicles for generalpurpose high-performance computing. Although a new Compute Unified Device Architecture (CUDA) programming model from NVIDIA offers improved programmability for general computing, programming GPGPUs is still complex and error-prone. This paper presents a compiler framework for automatic source-to-source translation of standard OpenMP applications into CUDA-based GPGPU applications. The goal of this translation is to further improve programmability and make existing OpenMP applications amenable to execution on GPGPUs. In this paper, we have identified several key transformation techniques, which enable efficient GPU global memory access, to achieve high performance. Experimental results from two important kernels (JACOBI and SPMUL) and two NAS OpenMP Parallel Benchmarks (EP and CG) show that the described translator and compile-time optimizations work well on both regular and irregular applications, leading to performance improvements of up to 50X over the unoptimized translation (up to 328X over serial on a CPU).

The analysis and the understanding of object manipulation scenarios based on computer vision techniques can be greatly facilitated if we can gain access to the full articulation of the manipulating hands and the 3D pose of the manipulated objects. Currently, there exist methods for tracking hands in interaction with objects whose 3D models are known. There are also methods that can reconstruct 3D models of objects that are partially observable in each frame of a sequence. However, to the best of our knowledge, no method can track hands in interaction with unknown objects. In this paper we propose such a method. Experimental results show that hand tracking can be achieved with an accuracy that is comparable to the one obtained by methods that assume knowledge of the object models. Additionally, as a by-product, the proposed method delivers accurate 3D models of the manipulated objects.

Todays commercial off-the-shelf computer systems are multicore computing systems as a combination of CPU, graphic processor (GPU) and custom devices. In comparison with CPU cores, graphic cards are capable to execute hundreds up to thousands compute units in parallel. To benefit from these GPU computing resources, applications have to be parallelized and adapted to the target architecture. In this paper we show our experience in applying the NQueens puzzle solution on GPUs using Nvidia's CUDA (Compute Unified Device Architecture) technology. Using the example of memory usage and memory access, we demonstrate that optimizations of CUDA programs may have contrary results on different CUDA architectures. Evaluation results will point out, that it is not sufficient to use new programming languages or compilers to achieve best results with emerging graphic card computing.

Current GPU computational power enables the execution of complex and parallel algorithms, such as Ray Tracing techniques supported by kD-trees for 3D scene rendering in real time. This work describes in detail the study and implementation of five different kD-Tree traversal algorithms using the parallel framework NVIDIA Compute Unified Device Architecture (CUDA), in order to point their pros and cons