• Journal of Computing Science and Engineering
• /
• v.11 no.2
• /
• pp.69-77
• /
• 2017

Recent GPUs have adopted cache memory to benefit general-purpose GPU (GPGPU) programs. However, unlike CPU programs, GPGPU programs typically have considerably less temporal/spatial locality. Moreover, the L1 data cache is used by many threads that access a data size typically considerably larger than the L1 cache, making it critical to bypass L1 data cache intelligently to enhance GPU cache performance. In this paper, we examine GPU cache access behavior and propose a simple hardware-based GPU cache bypassing method that can be applied to GPU applications without recompiling programs. Moreover, we introduce a hybrid method that integrates static profiling information and hardware-based bypassing to further enhance performance. Our experimental results reveal that hardware-based cache bypassing can boost performance for most benchmarks, and the hybrid method can achieve performance comparable to state-of-the-art compiler-based bypassing with considerably less profiling cost.

• International Journal of Advanced Culture Technology
• /
• v.4 no.4
• /
• pp.57-63
• /
• 2016

In this paper, we propose the optimized binarization in the GP-GPU. Because the binarinztion is esily paralledlized, we propose two ways of binary operations that utilize GP-GPU. The first method was to divide data load, subtraction and conversion, data store. The second method was processed collectibely. The second method was 2.52 times faster than the first method. After synthesizing the GP-GPU to the FPGA, the GP-GPU on the binarization were compared with the binarization on the ODROID XU. The binarization on the GP-GPU was 1.89 times faster than the binarization on the ODROID XU.

• 한국전산유체공학회:학술대회논문집
• /
• 2011.05a
• /
• pp.194-198
• /
• 2011

In this study, an efficient algorithm for Delaunay triangulation of a number of points which can be used on a GPU-based parallel computation is studied The developed algorithm is programmed using CUDA library. and the program takes full advantage of parallel computation which are concurrently performed on each of the threads on GPU. The results of partitioned triangulation collected from the GPU computation requires proper stitching between neighboring partitions and calculation of connectivities among triangular cells on CPU In this study, the effect of number of threads on the efficiency and total duration for Delaunay grid generation is studied. And it is also shown that GPU computing using CUDA for Delaunay grid generation is feasible and it saves total time required for the triangulation of the large number points compared to the sequential CPU-based triangulation programs.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2012.07a
• /
• pp.1-2
• /
• 2012

본 논문에서는 GPU 컴퓨팅 환경에서 물리적 모델링 기반의 음 합성 알고리즘을 수행하는 경우에 GPU의 개수에 따른 성능 및 에너지 효율의 변화를 분석한다. 실험결과, 6개의 GPU를 사용하였을 때 가장 좋은 성능을 보였으며, 1개의 GPU에서 가장 높은 에너지 효율을 보였다.

• ETRI Journal
• /
• v.45 no.6
• /
• pp.1035-1045
• /
• 2023

We propose a novel graphics processing unit (GPU) algorithm that can handle a large-scale 3D fast Fourier transform (i.e., 3D-FFT) problem whose data size is larger than the GPU's memory. A 1D FFT-based 3D-FFT computational approach is used to solve the limited device memory issue. Moreover, to reduce the communication overhead between the CPU and GPU, we propose a 3D data-transposition method that converts the target 1D vector into a contiguous memory layout and improves data transfer efficiency. The transposed data are communicated between the host and device memories efficiently through the pinned buffer and multiple streams. We apply our method to various large-scale benchmarks and compare its performance with the state-of-the-art multicore CPU FFT library (i.e., fastest Fourier transform in the West [FFTW]) and a prior GPU-based 3D-FFT algorithm. Our method achieves a higher performance (up to 2.89 times) than FFTW; it yields more performance gaps as the data size increases. The performance of the prior GPU algorithm decreases considerably in massive-scale problems, whereas our method's performance is stable.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.11 no.6
• /
• pp.325-334
• /
• 2016

Recently, Open Computing Language (OpenCL) has been proposed to provide a framework that supports heterogeneous computing platforms. By using an OpenCL framework, digital communication systems can support various protocols in a unified computing environment to achieve both high portability and high performance. This article introduces a parallel software decoder of Low Density Parity Check (LDPC) codes for China Multimedia Mobile Broadcasting (CMMB) on a heterogeneous platform. Each step of LDPC decoding has different parallelization characteristics. In this paper, steps suitable for task-level parallelization are executed on the CPU, and steps suitable for data-level parallelization are processed by the GPU. To improve the performance of the proposed OpenCL kernels for LDPC decoding operations, explicit thread scheduling, loop-unrolling, and effective data transfer techniques are applied. The proposed LDPC decoder achieves high performance by using heterogeneous multi-core processors on a unified computing framework.

• Journal of Electrical Engineering and Technology
• /
• v.13 no.1
• /
• pp.181-191
• /
• 2018

Rapid development of information technology makes our environment become smarter and massive high performance computers are providing powerful computing for that. Graphics Processing Unit (GPU) as a typical high performance component is being widely used for both graphics and general-purpose applications. Although it can greatly improve computing power, it also delivers significant power consumption and need sufficient power supplies. To make high performance computing more sustainable, the important step is to measure it. Current power technologies for GPU have some drawbacks, such as they are not applicable for power estimation at the early stage. In this article, we present a novel power technology to correlate power consumption and the characteristics at the programmer perspective, and then to estimate power consumption of source program without prerunning. We conduct experiments on Nvidia's GT740 platform; the results show that our power model is more accurately than regression model and has an average error of 2.34% and the maximum error of 9.65%.

• Journal of the Korea Computer Graphics Society
• /
• v.17 no.3
• /
• pp.9-19
• /
• 2011

Computing triangle-triangle intersections has been a fundamental task required for many 3D geometric problems. We propose a novel robust GPU algorithm to efficiently compute intersections in a large triangle set. The algorithm has three stages:k-d tree construction, triangle pair generation, and exact intersection computation. All three stages are executed on GPU except, for unsafe triangle pairs. Unsafe triangle pairs are robustly handled by CLP(controlled linear perturbation) on a CPU thread. They are identified by floating-point filtering while exact intersection is computed on GPU. Many triangles crossing a split plane are duplicated in k-d tree construction, which form a lot of redundant triangle pairs later. To eliminate them efficiently, we use a split index which can determine redundancy of a pair by a simple bitwise operation. We applied the proposed algorithm to computing 3D Minkowski sum boundaries to verify its efficiency and robustness.

• Journal of Internet Computing and Services
• /
• v.18 no.3
• /
• pp.29-36
• /
• 2017

In these days, 3D dynamic simulation is closely related to many industries. In the past, physically-based 3D simulation was used mainly in the car crash or construction related fields, but it also plays an important role in movies or games today. Many mathematical computations are needed to represent the 3D object realistically, but it is difficult to process a large amount of calculations for simulation of application based on CPU in real-time. Recently, with the advanced graphic hardware and improved architecture, GPU can be utilized for the general purposes of computation function as well as graphic computation. Many approaches using GPU have been applied for various research fields. In this paper, we analyze the performance variation of two cloth simulation algorithms based on GPU according to the change of execution properties of GPU shaders in oder to optimize the performance of GPU-based cloth simulation. Cloth simulation is implemented by the spring centric algorithm and node centric algorithm with GPU parallel computing using compute shader of GLSL 4.3. We compare the performance of between these algorithms according to the change of the size and dimension of work group. The experiment is repeated to 10 times during 5,000 frames for each test and experimental results are provided by averaging of FPS. The experimental result shows that the node centric algorithm is executed in higher speed than the spring centric algorithm.

• KIPS Transactions on Computer and Communication Systems
• /
• v.7 no.4
• /
• pp.99-102
• /
• 2018

Caffe is a neural net learning software which is widely used in academic researches. The GPU memory capacity is one of the most important aspects of designing neural net architectures. For example, many object detection systems require to use less than 12GB to fit a single GPU. In this paper, we extended Caffe to allow to use more than 12GB GPU memory. To verify the effectiveness of the extended software, we executed some training experiments to determine the learning efficiency of the object detection neural net software using a PC with three GPUs.