Search results for: CUDA.

Authors: Sanghyeok Oh, Keechul Jung

Abstract:

Although lots of research work has been done for human pose recognition, the view-point of cameras is still critical problem of overall recognition system. In this paper, view-point insensitive human pose recognition is proposed. The aims of the proposed system are view-point insensitivity and real-time processing. Recognition system consists of feature extraction module, neural network and real-time feed forward calculation. First, histogram-based method is used to extract feature from silhouette image and it is suitable for represent the shape of human pose. To reduce the dimension of feature vector, Principle Component Analysis(PCA) is used. Second, real-time processing is implemented by using Compute Unified Device Architecture(CUDA) and this architecture improves the speed of feed-forward calculation of neural network. We demonstrate the effectiveness of our approach with experiments on real environment.

Keywords: computer vision, neural network, pose recognition, view-point insensitive, PCA, CUDA.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1339

Authors: Tadayoshi Horita, Masakazu Akiba, Mina Terauchi, Tsuneo Kanno

Abstract:

This paper gives a consideration of the achievement of productive level parallel programming skills, based on the data of the graduation studies in the Polytechnic University of Japan. The data show that most students can achieve only parallel programming skills during the graduation study (about 600 to 700 hours), if the programming environment is limited to GPGPUs. However, the data also show that it is a very high level task that a student achieves productive level parallel programming skills during only the graduation study. In addition, it shows that the parallel programming environments for GPGPU, such as CUDA and OpenCL, may be more suitable for parallel computing education than other environments such as MPI on a cluster system and Cell.B.E. These results must be useful for the areas of not only software developments, but also hardware product developments using computer technologies.

Keywords: Parallel computing, programming education, GPU, GPGPU, CUDA, OpenCL, MPI, Cell.B.E.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1687

Authors: Irakli V. Gugushvili, Nickolay M. Evstigneev

Abstract:

Numerical integration of initial boundary problem for advection equation in 3 ℜ is considered. The method used is  conditionally stable semi-Lagrangian advection scheme with high order interpolation on unstructured mesh. In order to increase time step integration the BFECC method with limiter TVD correction is used. The method is adopted on parallel graphic processor unit environment using NVIDIA CUDA and applied in Navier-Stokes solver. It is shown that the calculation on NVIDIA GeForce 8800  GPU is 184 times faster than on one processor AMDX2 4800+ CPU. The method is extended to the incompressible fluid dynamics solver. Flow over a Cylinder for 3D case is compared to the experimental data.

Keywords: Advection equations, CUDA technology, Flow overthe 3D Cylinder, Incompressible Pressure Projection Solver, Parallel computation.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2846

Authors: Jie Zhao, Meng Su

Abstract:

Image recognition enables machine-like robotics to understand a scene and plays an important role in computer vision applications. Computer vision platforms as physical infrastructure, supporting Neural Networks for image recognition, are deterministic to leverage the performance of different Neural Networks. In this paper, three different computer vision platforms – edge AI (Jetson Nano, with 4GB), a standalone laptop (with RTX 3000s, using CUDA), and a web-based device (Google Colab, using GPU) are investigated. In the case study, four prominent neural network architectures (including AlexNet, VGG16, GoogleNet, and ResNet (34/50)), are deployed. By using public ImageNets (Cifar-10), our findings provide a nuanced perspective on optimizing image recognition tasks across Edge-AI platforms, offering guidance on selecting appropriate neural network structures to maximize performance under hardware constraints.

Keywords: AlexNet, VGG, GoogleNet, ResNet, ImageNet, Cifar-10, Edge AI, Jetson Nano, CUDA, GPU.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 220

Authors: Rossella Arcucci, Luisa D’Amore, Simone Celestino, Giuseppe Scotti, Giuliano Laccetti

Abstract:

This work is the first dowel in a rather wide research activity in collaboration with Euro Mediterranean Center for Climate Changes, aimed at introducing scalable approaches in Ocean Circulation Models. We discuss designing and implementation of a parallel algorithm for solving the Variational Data Assimilation (DA) problem on Graphics Processing Units (GPUs). The algorithm is based on the fully scalable 3DVar DA model, previously proposed by the authors, which uses a Domain Decomposition approach (we refer to this model as the DD-DA model). We proceed with an incremental porting process consisting of 3 distinct stages: requirements and source code analysis, incremental development of CUDA kernels, testing and optimization. Experiments confirm the theoretic performance analysis based on the so-called scale up factor demonstrating that the DD-DA model can be suitably mapped on GPU architectures.

Keywords: Data Assimilation, Parallel Algorithm, GPU architectures, Ocean Models.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 2010

Authors: Mohammad Amin Safi, Mahmud Ashrafizaadeh, Amir Ali Ashrafizaadeh

Abstract:

A highly optimized implementation of binary mixture diffusion with no initial bulk velocity on graphics processors is presented. The lattice Boltzmann model is employed for simulating the binary diffusion of oxygen and nitrogen into each other with different initial concentration distributions. Simulations have been performed using the latest proposed lattice Boltzmann model that satisfies both the indifferentiability principle and the H-theorem for multi-component gas mixtures. Contemporary numerical optimization techniques such as memory alignment and increasing the multiprocessor occupancy are exploited along with some novel optimization strategies to enhance the computational performance on graphics processors using the C for CUDA programming language. Speedup of more than two orders of magnitude over single-core processors is achieved on a variety of Graphical Processing Unit (GPU) devices ranging from conventional graphics cards to advanced, high-end GPUs, while the numerical results are in excellent agreement with the available analytical and numerical data in the literature.

Keywords: Lattice Boltzmann model, Graphical processing unit, Binary mixture diffusion, 2D flow simulations, Optimized algorithm.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 1556

Authors: Sergey Kopysov, Nikita Nedozhogin, Leonid Tonkov

Abstract:

The article presents a parallel iterative solver for large sparse linear systems which can be used on a heterogeneous platform. Traditionally, the problem of solving linear systems do not scale well on cluster containing multiple Central Processing Units (multi-CPUs cluster) or cluster containing multiple Graphics Processing Units (multi-GPUs cluster). For example, most of the attempts to implement the classical conjugate gradient method were at best counted in the same amount of time as the problem was enlarged. The paper proposes the pipelined variant of the conjugate gradient method (PCG), a formulation that is potentially better suited for hybrid CPU/GPU computing since it requires only one synchronization point per one iteration, instead of two for standard CG (Conjugate Gradient). The standard and pipelined CG methods need the vector entries generated by current GPU and other GPUs for matrix-vector product. So the communication between GPUs becomes a major performance bottleneck on miltiGPU cluster. The article presents an approach to minimize the communications between parallel parts of algorithms. Additionally, computation and communication can be overlapped to reduce the impact of data exchange. Using pipelined version of the CG method with one synchronization point, the possibility of asynchronous calculations and communications, load balancing between the CPU and GPU for solving the large linear systems allows for scalability. The algorithm is implemented with the combined use of technologies: MPI, OpenMP and CUDA. We show that almost optimum speed up on 8-CPU/2GPU may be reached (relatively to a one GPU execution). The parallelized solver achieves a speedup of up to 5.49 times on 16 NVIDIA Tesla GPUs, as compared to one GPU.

Keywords: Conjugate Gradient, GPU, parallel programming, pipelined algorithm.

Procedia APA BibTeX Chicago EndNote Harvard JSON MLA RIS XML ISO 690 PDF Downloads 371