METHOD OF ACCELERATION OF NEURAL NETWORK TASKS ON HETEROGENEOUS CPU-GPU SYSTEMS ; МЕТОД ПРИСКОРЕННЯ ВИКОНАННЯ ЗАДАЧ НЕЙРОННИХ МЕРЕЖ НА ГЕТЕРОГЕННИХ СИСТЕМАХ CPU-GPU
In: Технічні науки та технології; № 2(24) (2021): Технічні науки та технології; 131-140 ; Technical sciences and technology, 2021
academicJournal
Zugriff:
Research areas related to heterogeneous computing are currently underdeveloped. In machine learning, huge efforts are directed towards creating more efficient algorithms and getting more performance from GPUs. There is also a growing demand for the use of neural networks in many practical areas, since many large companies are developing intelligent systems that work directly with Big Data solutions. General purpose of GPU computing hands GPUs time-consuming tasks, however there is limited research on the capabilities of heterogeneous systems. Neural networks as a way to find solutions to practical problems are becoming more widespread every day. In this regard, there is a demand for high-performance computer systems. Articles dealing with heterogeneous systems provide promising results. Heterogeneous CPU-GPU systems, although not well understood, have been shown to speed up tasks associated with image processing and big data tasks. Most modern machine learning articles and libraries perform tasks on CPUs or GPUs, however, heterogeneous systems have not been sufficiently investigated for possibilities in machine learning tasks. The aim of the study is to develop an algorithm for scheduling a problem for a heterogeneous system and to carry out a comparative analysis of the results on the use of homogeneous systems. An algorithm for scheduling a problem for a heterogeneous system based on a linear regression model is described, the problem which is modeled is shown and a comparative analysis of the results of various heterogeneous systems is carried out. The results show that the heterogeneous approach is effective in neural network tasks. From the results we can conclude that the program takes advantage of new processors and GPU performance acceleration in the range of 1.11 to 4.39, as well as CPU performance acceleration in the range of 0.96 to 3.48 show that in most cases, CPU performance can be improved if the GPU is used at the same time. ; У статті розглянуто питання використання гетерогенних систем CPU-GPU ...
Titel: |
METHOD OF ACCELERATION OF NEURAL NETWORK TASKS ON HETEROGENEOUS CPU-GPU SYSTEMS ; МЕТОД ПРИСКОРЕННЯ ВИКОНАННЯ ЗАДАЧ НЕЙРОННИХ МЕРЕЖ НА ГЕТЕРОГЕННИХ СИСТЕМАХ CPU-GPU
|
---|---|
Autor/in / Beteiligte Person: | Русінов , Володимир ; Череватенко, Олексій ; Пустовіт, Леонід ; Пустовіт, Олександр |
Link: | |
Zeitschrift: | Технічні науки та технології; № 2(24) (2021): Технічні науки та технології; 131-140 ; Technical sciences and technology, 2021 |
Veröffentlichung: | Національний університет «Чернігівська політехніка», 2021 |
Medientyp: | academicJournal |
Schlagwort: |
|
Sonstiges: |
|