TITLE

A comparative study of GPU programming models and architectures using neural networks

AUTHOR(S)
Pallipuram, Vivek; Bhuiyan, Mohammad; Smith, Melissa
PUB. DATE
September 2012
SOURCE
Journal of Supercomputing;Sep2012, Vol. 61 Issue 3, p673
SOURCE TYPE
Academic Journal
DOC. TYPE
Article
ABSTRACT
Recently, General Purpose Graphical Processing Units (GP-GPUs) have been identified as an intriguing technology to accelerate numerous data-parallel algorithms. Several GPU architectures and programming models are beginning to emerge and establish their niche in the High-Performance Computing (HPC) community. New massively parallel architectures such as the Nvidia's Fermi and AMD/ATi's Radeon pack tremendous computing power in their large number of multiprocessors. Their performance is unleashed using one of the two GP-GPU programming models: Compute Unified Device Architecture (CUDA) and Open Computing Language (OpenCL). Both of them offer constructs and features that have direct bearing on the application runtime performance. In this paper, we compare the two GP-GPU architectures and the two programming models using a two-level character recognition network. The two-level network is developed using four different Spiking Neural Network (SNN) models, each with different ratios of computation-to-communication requirements. To compare the architectures, we have chosen the two extremes of the SNN models for implementation of the aforementioned two-level network. An architectural performance comparison of the SNN application running on Nvidia's Fermi and AMD/ATi's Radeon is done using the OpenCL programming model exhausting all of the optimization strategies plausible for the two architectures. To compare the programming models, we implement the two-level network on Nvidia's Tesla C2050 based on the Fermi architecture. We present a hierarchy of implementations, where we successively add optimization techniques associated with the two programming models. We then compare the two programming models at these different levels of implementation and also present the effect of the network size (problem size) on the performance. We report significant application speed-up, as high as 1095× for the most computation intensive SNN neuron model, against a serial implementation on the Intel Core 2 Quad host. A comprehensive study presented in this paper establishes connections between programming models, architectures and applications.
ACCESSION #
78437661

 

Related Articles

  • Multi-robot task allocation using CNP combines with neural network. Yuan, Quande; Guan, Yi; Hong, Bingrong; Meng, Xiangping // Neural Computing & Applications;Dec2013, Vol. 23 Issue 7/8, p1909 

    Contract Net Protocol is a suitable method for multi-robot task allocation problems. However, it is difficult to find a function to evaluate robots’ bids when each robot gives more than one bid price to reflect its different abilities. We propose a method to fuse these prices and to...

  • Guest Editor's Introduction to the Special Issue On Neural Network Software and Systems. Gelenbe, Erol // IEEE Transactions on Software Engineering;Jul92, Vol. 18 Issue 7, p549 

    This guest editorial introduces a special issue of the periodical "IEEE Transactions on Software Engineering" which focuses on neural network software and systems. It outlines the history of artificial neural networks and their role in the origin of the theory of computing. The special issue...

  • Divide and Conquer Approach in Reducing ANN Training Time for Small and Large Data. Mohamad, Mumtazimah; Saman, Md Yazid Mohd; Hitam, Muhammad Suzuri // Journal of Applied Sciences;1/1/2013, Vol. 13 Issue 1, p133 

    Artificial Neural Networks (ANN) are able to simplify recognition tasks and have been steadily improving both in accuracy and efficiency. Classical ANN, as a universal approximator, has been proven to be a more versatile and flexible method compared to modern, high-end algorithms. However, there...

  • Application of Neural Network Technologies for Price Forecasting in the Liberalized Electricity Market. Gerikh, Valentin P.; Kolosok, Irina N.; Kurbatsk, Victor G.; Tomin, Nikita V. // Power & Electrical Engineering;2009, Issue 25, preceding p91 

    The paper presents the results of experimental studies concerning calculation of electricity prices in different price zones in Russia and Europe. The calculations are based on the intelligent software "ANAPRO" that implements the approaches based on the modern methods of data analysis and...

  • A New Method for Cardiovascular Disease Clinical Diagnosis Based on Artificial Neural Network Model. Huang Zhao-Ming; Zeng Xue-Mei // Information Technology Journal;2013, Vol. 12 Issue 21, p6277 

    Diagnosis. In order to improve the accuracy of Clinical Diagnosis for Cardiovascular Disease, ANN (Artificial Neural Network is introduced in this paper. 200 cases of cardiovascular disease which have similar symptom and different diagnosis are sampled from our database. BP Network model in...

  • Proposition of a Geotechnical Mapping Based on Artificial Neural Networks for the Town of Caucaia, Ceará, Brazil for Paving Purposes. Alves Ribeiro, Antonio Júnior; da Silva, Carlos Augusto Uchôa; Araújo Barroso, Suelly Helena de // International Journal of Engineering & Technology;Oct2012, Vol. 12 Issue 5, p65 

    This research focuses on the development of a method, based on Artificial Neural Networks (ANN), aimed to infer the geotechnical characteristics of the subgrade of Caucaia Town, located in the metropolitan region of Fortaleza - Ceará, Brazil based on biophysical variables (pedology, geology,...

  • Neural-Network Computers.  // Futurist;Sep/Oct89, Vol. 23 Issue 5, p56 

    Describes the development of neural-network computers that are self-learning with self-adaptive capabilities. Current uses of neural network computers, including telecommunications, risk analysis and robotics; Interest of the military in using neural networks to solve tactical problems; Ways in...

  • The forecasting research of Beijing tourism demand based on the BP neural network. Yang Yu; Shimin Wang // Applied Mechanics & Materials;2014, Issue 571-572, p128 

    This paper describes the basic principles and algorithm of the BP neural network and builds a forecasting model of Beijing tourism demand based on the BP neural network. The forecasting model can forecast and analyze the number of tourists in Beijing in the future, which using the MATLAB tools...

  • Scaling analysis of a neocortex inspired cognitive model on the Cray XD1. Rice, Kenneth; Taha, Tarek; Vutsinas, Christopher // Journal of Supercomputing;Jan2009, Vol. 47 Issue 1, p21 

    This paper presents the implementation and scaling of a neocortex inspired cognitive model on a Cray XD1. Both software and reconfigurable logic based FPGA implementations of the model are examined. This model belongs to a new class of biologically inspired cognitive models. Large scale versions...

Share

Read the Article

Courtesy of THE LIBRARY OF VIRGINIA

Sorry, but this item is not currently available from your library.

Try another library?
Sign out of this library

Other Topics