GPU Computing Taxonomy

Osman, Abdelrahman Ahmed Mohamed

doi:10.5772/intechopen.68179

Cited by 3 publications

(4 citation statements)

References 2 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Table 2, adapted from Su et al (2013) and Karimi et al (2010), briefly presents the different concepts used in OpenCL (Wang et al, 2017) and CUDA (Su et al, 2013;Karimi et al, 2010). Regardless, there are imperfection elements (Osman, 2017) of GPU architecture to the DNN-dominant tasks.…”

Section: Machine Learningmentioning

confidence: 99%

“…Regardless, there are imperfection elements (Osman, 2017) of GPU architecture to the DNN-dominant tasks.…”

Section: Hardware Design and Optimization Techniquesmentioning

confidence: 99%

“…Second, the users implement the model with the APIs supporting virtual machines without knowing hidden GPU architectures. This concept and the absence of detailed application may interfere with manipulating the best performance out of the GPUs (Osman, 2017). For example, the users would not know how to reuse the memory under the APIs without cache model visibility.…”

Section: Hardware Design and Optimization Techniquesmentioning

confidence: 99%

“…Furthermore, 64-bit floating- point numbers are limited in the programmable parts of GPUs, where the CPUs support those primitive types. The evolution of GPUs will concentrate on built-in support for diverse and higher floating-point numbers gradually (Osman, 2017). Last but not least, GPU architecture possesses several mini graphics processors, with their computation unit and local cache fitting the matrix multiplication only.…”

Section: Hardware Design and Optimization Techniquesmentioning

confidence: 99%

See 3 more Smart Citations

A review of in-memory computing for machine learning: architectures, options

Snasel,

Dang,

Kueng

et al. 2023

IJWIS

View full text Add to dashboard Cite

Purpose This paper aims to review in-memory computing (IMC) for machine learning (ML) applications from history, architectures and options aspects. In this review, the authors investigate different architectural aspects and collect and provide our comparative evaluations. Design/methodology/approach Collecting over 40 IMC papers related to hardware design and optimization techniques of recent years, then classify them into three optimization option categories: optimization through graphic processing unit (GPU), optimization through reduced precision and optimization through hardware accelerator. Then, the authors brief those techniques in aspects such as what kind of data set it applied, how it is designed and what is the contribution of this design. Findings ML algorithms are potent tools accommodated on IMC architecture. Although general-purpose hardware (central processing units and GPUs) can supply explicit solutions, their energy efficiencies have limitations because of their excessive flexibility support. On the other hand, hardware accelerators (field programmable gate arrays and application-specific integrated circuits) win on the energy efficiency aspect, but individual accelerator often adapts exclusively to ax single ML approach (family). From a long hardware evolution perspective, hardware/software collaboration heterogeneity design from hybrid platforms is an option for the researcher. Originality/value IMC’s optimization enables high-speed processing, increases performance and analyzes massive volumes of data in real-time. This work reviews IMC and its evolution. Then, the authors categorize three optimization paths for the IMC architecture to improve performance metrics.

show abstract

Section: Machine Learningmentioning

confidence: 99%

“…Regardless, there are imperfection elements (Osman, 2017) of GPU architecture to the DNN-dominant tasks.…”

Section: Hardware Design and Optimization Techniquesmentioning

confidence: 99%

Section: Hardware Design and Optimization Techniquesmentioning

confidence: 99%

Section: Hardware Design and Optimization Techniquesmentioning

confidence: 99%

See 2 more Smart Citations

A review of in-memory computing for machine learning: architectures, options

Snasel,

Dang,

Kueng

et al. 2023

IJWIS

View full text Add to dashboard Cite

show abstract

In-Memory Computing Architectures for Big Data and Machine Learning Applications

Snåšel¹,

Dang²,

Phạm³

et al. 2022

Future Data and Security Engineering. Big Data, Security and Privacy, Smart City and Industry 4.0 Applications

View full text Add to dashboard Cite

Model of a Parallel-Pipeline Computational Process for Solving a System of Grid Equations

Litvinov,

Rudenko,

Gracheva

2023

Vestnik Donskogo gosudarstvennogo tehničeskogo universiteta

View full text Add to dashboard Cite

Introduction. Environmental problems arising in shallow waters and caused by both natural and man-made factors annually do significant damage to aquatic systems and coastal territories. It is possible to identify these problems in a timely manner, as well as ways to eliminate them, using modern computing systems. But earlier studies have shown that the resources of computing systems using only a central processor are not enough to solve large scientific problems, in particular, to predict major environmental accidents, assess the damage caused by them, and determine the possibilities of their elimination. For these purposes, it is proposed to use models of the computing system and decomposition of the computational domain to develop an algorithm for parallel-pipeline calculations. The research objective was to create a model of a parallel-conveyor computational process for solving a system of grid equations by a modified alternating-triangular iterative method using the decomposition of a three-dimensional uniform computational grid that takes into account technical characteristics of the equipment used for calculations.Materials and Methods. Mathematical models of the computer system and computational grid were developed. The decomposition model of the computational domain was made taking into account the characteristics of a heterogeneous system. A parallel-pipeline method for solving a system of grid equations by a modified alternating-triangular iterative method was proposed.Results. A program was written in the CUDA C language that implemented a parallel-pipeline method for solving a system of grid equations by a modified alternating-triangular iterative method. The experiments performed showed that with an increase in the number of threads, the computation time decreased, and when decomposing the computational grid, it was rational to split into fragments along coordinate z by a value not exceeding 10. The results of the experiments proved the efficiency of the developed parallel-pipeline method.Discussion and Conclusion. As a result of the research, a model of a parallel-pipeline computing process was developed using the example of one of the most time-consuming stages of solving a system of grid equations by a modified alternating-triangular iterative method. Its construction was based on decomposition models of a three-dimensional uniform computational grid, which took into account the technical characteristics of the equipment used in the calculations. This program can provide you for the acceleration of the calculation process and even loading of program flows in time. The conducted numerical experiments validated the mathematical model of decomposition of the computational domain.

show abstract

GPU Computing Taxonomy

Cited by 3 publications

References 2 publications

A review of in-memory computing for machine learning: architectures, options

A review of in-memory computing for machine learning: architectures, options

In-Memory Computing Architectures for Big Data and Machine Learning Applications

Model of a Parallel-Pipeline Computational Process for Solving a System of Grid Equations

Contact Info

Product

Resources

About