Wyniki wyszukiwania - Biblioteka Nauki

Nowa wersja platformy, zawierająca wyłącznie zasoby pełnotekstowe, jest już dostępna.
Przejdź na https://bibliotekanauki.pl

Ograniczanie wyników

Znaleziono wyników: 3

Liczba wyników na stronie

Wyniki wyszukiwania

Wyszukiwano:
w słowach kluczowych: high-performance computing

Sortuj według:

Ogranicz wyniki do:

High-performance simulation-based algorithms for an alpine ski racer's trajectory optimization in heterogeneous computer systems

100%

Dębski R.

tom 24

nr 3

551-566

Effective, simulation-based trajectory optimization algorithms adapted to heterogeneous computers are studied with reference to the problem taken from alpine ski racing (the presented solution is probably the most general one published so far). The key idea behind these algorithms is to use a grid-based discretization scheme to transform the continuous optimization problem into a search problem over a specially constructed finite graph, and then to apply dynamic programming to find an approximation of the global solution. In the analyzed example it is the minimum-time ski line, represented as a piecewise-linear function (a method of elimination of unfeasible solutions is proposed). Serial and parallel versions of the basic optimization algorithm are presented in detail (pseudo-code, time and memory complexity). Possible extensions of the basic algorithm are also described. The implementation of these algorithms is based on OpenCL. The included experimental results show that contemporary heterogeneous computers can be treated as μ-HPC platforms-they offer high performance (the best speedup was equal to 128) while remaining energy and cost efficient (which is crucial in embedded systems, e.g., trajectory planners of autonomous robots). The presented algorithms can be applied to many trajectory optimization problems, including those having a black-box represented performance measure.

Machine Learning and High-Performance Computing Hybrid Systems, a New Way of Performance Acceleration in Engineering and Scientific Applications

72%

Gepner P.

2021

tom Vol. 25

27--36

Machine learning is one of the hottest topics in IT industry as well as in academia. Some of the IT leaders and scientists believe that this is going to totally revolutionise the industry. This transformation is happening on both fronts, one is the application and software paradigm, the other is at the hardware and system level. At the same time, the High-Performance Computing segment is striving to achieve the level of Exascale performance. It is not debatable that to meet such level of performance and keep the cost of system and power consumption on reasonable level is not a trivial task. In this article, we try to look at a potential solution to these problems and discuss a new approach to building systems and software to meet these challenges and the growing needs of the computing power for HPC systems on the one hand, but also be ready for a new type of workload including Artificial Intelligence type of applications.

Performance enhancement of CUDA applications by overlapping data transfer and Kernel execution

72%

Raju K. , Chiplunkar N. N.

Applied Computer Science

2021

tom Vol. 17, no 3

5--18

The CPU-GPU combination is a widely used heterogeneous computing system in which the CPU and GPU have different address spaces. Since the GPU cannot directly access the CPU memory, prior to invoking the GPU function the input data must be available on the GPU memory. On completion of GPU function, the results of computation are transferred to CPU memory. The CPU-GPU data transfer happens through PCIExpress bus. The PCI-E bandwidth is much lesser than that of GPU memory. The speed at which the data is transferred is limited by the PCI-E bandwidth. Hence, the PCI-E acts as a performance bottleneck. In this paper two approaches are discussed to minimize the overhead of data transfer, namely, performing the data transfer while the GPU function is being executed and reducing the amount of data to be transferred to GPU. The effectiveness of these approaches on the execution time of a set of CUDA applications is realized using CUDA streams. The results of our experiments show that the execution time of applications can be minimized with the proposed approaches.