Papers by Keyword: GPU

Abstract: This paper introduces a procedure for parallel computing with the Dynamic Relaxation method (DR) on a Graphic Processing Unit (GPU).This method facilitates the consideration of a variety of nonlinearities in an easy and explicit manner.Because of the presence of inertial forces, a static problem leads to a transient dynamic problem where the Central Difference Method is usedas a method for direct integration of equations of motion which arise from the Finite Element model.The natural characteristic of this explicit method is that the scheme can be easily parallelized. The assembly of a global stiffness matrix is not required.Due to slow convergence with this method, the high performance which GPUs provide is strongly suitable for this kind of computation.NVIDIA's CUDA is used for general-purpose computing on graphics processing units (GPGPU) for NVIDIA's GPUs with CUDA capability.

Abstract: There are various vendors of FFT libraries, but there is no software available for it automatic benchmarking on all available devices. In this article an application that allows easy measure the performance and precision of various FFT libraries on the available GPUs and CPUs is presented. This application has been used to find out the fastest FFT library for NVIDIA GTX TESLA and NVIDIA GTX TITAN. The obtained results shown that the best implementation is provided by cuFFT library developed by NVIDIA.

Abstract: Defective products are unavoidable in printed circuit board production process, so rapid detection and identificationmethods are badly in need of. PCB surface defect detection including a series of processing such as surface imagecapture, mixed noise filtering,images registering and so on, so it takes a lot of CPU time. To improve detection speed, based on GPU parallel computing platform, we designed a reasonable parallel processing system for PCB defect detectionto meet the need of real-time requirements of a production line. Experimental results show that parallel image processing algorithms based on GPU can achieve good results compared to the CPU-based serial algorithm (with speed up ratio up to8.34 in this paper),providing a new approachfor rapid detection of PCB surface defect.

Abstract: The article presents a comparison of the performance of multiple rendering engines for the Blender program. Rendering times for an example model were compared in this work. Both CPU and GPU execution modes were included, using the older, more economical cards based on the Fermi architecture (GeForce GT 440 in this case) and the newer, more efficient based on Kepler architecture (GeForce GTX 680).

Abstract: Mild-slope equation is usually used in many simulation applications. The equation has obviously benefit which based on physical method that can present the real status of water, but the shortcoming is also clearly that the calculations spending lots of times which not support some real-time applications. We use hyperbola to simple the equation calculation process, and use alternating directions implicit (ADI) way to solve this equation. The result shows that the ADI way can use less calculation and less time to accomplish the calculation. We also use the new features of GPU(graphics process unit) to speed up the calculation so that we can render the surface in real-time application.

Abstract: We present a parallel solution based on CUDA for accelerating the computation for solving large-scale Finite Element equations in electrical and magnetic field. JCG is used for solving equations and corresponding kernel function is designed for spMV. A computation speed test for solving FE equations is taken on NVIDIA Tesla K20c GPU hardware platform, the result proves that the method of kernel can reach 17.1 times faster than the solution using CPU, however it cannot ensure the advantage with CPU if we only use the lib functions on GPU to solve equations.

Abstract: In this study, the Auxiliary Power Unit (APU) and the Ground Power Unit (GPU) that are required meet the energy demand in ground operations of aircraft were compared in terms of their damages –human health (disability adjusted life years- DALYs), ecosystem quality (PDF.m².yr) and resources (MJ Surplus). These damages were calculated with gate-to-gate method of Life Cycle Assessment (LCA). This assessment was held on three main stages; capital investment, operation and maintenance. SimaPro 7.2.4 was used for this calculation. Human Health Damages (HHDs) of APU were found to be (0.0557 -disability adjusted life years- DALYs) 20.34 days/year while HHDs of GPU were calculated as (0.0411 -disability adjusted life years-DALYs) 15 days/year. While Ecosystem Quality Damages (EQDs) of APU were calculated as 103366.37 PDF.m².yr, EQDs of GPU were found to be 82651.23 PDF.m².yr. In other words, although the use of APU gives rise to loss of 0.1 species –1 year, 1 square kilometer area, the use of GPU gives rise to loss of 0.083 species. With regards to Resources Damages (RDs), 62125 MJ surplus energy needed for future extractions of minerals and fossil fuel considering APU, one the other hand 36002 MJ surplus energy needed for future considering GPU. As a result, the use of APU effects on human health, ecosystem quality and resources was determined greater than the use of GPU for 30 minute. The use of GPU has less impact on human health, ecosystem quality and resources in ground operations.

Abstract: In artificial society simulation, each artificial population needs road path planning in the process of travel. However, because of the large amounts of populations in artificial society, road path planning will cost lots computational resources and time, thus this process has terrible efficiency to the performance of the simulation system. In order to solve this problem, this article firstly makes use of CPU to generate the artificial populations, travel logs, and construct the road network models; then computes the shortest road path between each two environments and load the results in RAM for prepare; lastly, sends the ID and its start point and destination of the population who need road path querying to GPU at current simulation time in the simulation process, and then takes advantages of GPU to query the road path and return the results back. In this way, we can obviously reduce the time costs in the process of road path querying and enormously improve the performance of the whole simulation system.

Abstract: In this paper, frame differential method will be used to detect moving targets in a static background video file, and pre-contour can be obtained by binarizing the detected targets . However,, the result are not what was expected, so the expansion and corrosion of mathematical morphology are used to extract the final contour of moving targets. in the progress of dealing with massive data , mathematical morphology is not good enough to achieve the need of the real-time in video surveillance. Considering the dilation and erosion is a kind of parallel processing operations, in order to improve the speed of mathematical morphology operations, this paper offers detailed implementation process of the dilation algorithm for parallel computing on GPU. Experimental results showed that GPU parallel processing on mathematical morphology algorithm faster than the CPU serial processing.

Abstract: For spreading capture large computation and computational problems of slow, capture method is proposed based on GPU acceleration, the conversion cycle-related acquisition algorithm based on CUDA thread blocks for the implementation of the process, so spreading the capture process is completely in the GPU to accelerate implementation, achieve better results in the calculation, while significantly improving the computational speed of operation. Experimental results show that GPU-based capture method effectively improve the efficiency of the system, compared to CPU operation speed is increased by about 42 times.