![International Journal of Trend in Scientific Research and Development (IJTSRD)
Volume 4 Issue 1, December 2019 Available Online: www.ijtsrd.com e-ISSN: 2456 – 6470
@ IJTSRD | Unique Paper ID – IJTSRD29648 | Volume – 4 | Issue – 1 | November-December 2019 Page 705
GPU Computing: An Introduction
Matthew N. O. Sadiku, Adedamola A. Omotoso, Sarhan M. Musa
Roy G. Perry College of Engineering, Prairie View A&M University, Prairie View, Texas
ABSTRACT
The graphics processing unit (GPU) is a computer chip that performs rapid
mathematical calculations. GPU is a ubiquitous device which appears inevery
computing systems such PC, laptop, desktop, and workstation. It is a many-
core multithreaded multiprocessor that excels at both graphics and non-
graphic applications. GPU computing is using a GPU as a co-processor to
accelerate CPUs scientific and engineering computing. The paper provides a
brief introduction to GPU computing.
KEYWORDS:graphicsprocessingunit, GPUcomputing, heterogeneouscomputing,
hybrid computing
How to cite this paper: Matthew N. O.
Sadiku | Adedamola A. Omotoso | Sarhan
M. Musa "GPU Computing: An
Introduction"
Published in
International Journal
of Trend in Scientific
Research and
Development(ijtsrd),
ISSN: 2456-6470,
Volume-4 | Issue-1,
December 2019, pp.705-706, URL:
www.ijtsrd.com/papers/ijtsrd29648.pdf
Copyright © 2019 by author(s) and
International Journal ofTrendinScientific
Research and Development Journal. This
is an Open Access article distributed
under the terms of
the Creative
CommonsAttribution
License (CC BY 4.0)
(http://creativecommons.org/licenses/by
/4.0)
INRODUCTION
The efficiency of anycomputersimulationsdependsonthree
factors [1]: the formulation of the theory describing the
process, numerical methods employed, and the hardware
capabilities. Graphics Processing Unit (GPU) computing is
regarded today as the most powerful computational
hardware. GPU is not the only type of accelerator core that
has gained interest recently. Other include field
programmable gate array (FPGA) and the Cell Broadband
Engine (Cell BE), which receive less attention [2].
Originally, GPU was conceived and developed for rendering
graphics. However, due to it high performance and low cost,
it has become the new standard of non-graphic application
such as image processing, image restoration, filtering,
interpolation, and reconstruction. It has become an
indispensable part of today's computing systems. In recent
years, substantial efforts were made to adapt many
algorithms for massively-parallel GPU-based systems since
the GPU can perform many calculations simultaneously.
GPU computing is the application of a GPU to do general
purpose scientific and engineering computing. Central
Processing Units (CPUs) are task-parallel processors, while
GPUs are data-parallel. GPU computing is not replacing CPU
computing. Each approach has its own advantages and
limitations. A GPU is used along with a CPU to accelerate
scientific and engineering applications.Forthisreason,CPU-
GPU computing is also known as heterogeneous computing
or hybrid computing.
GPU BASICS
The term GPU was popularized by NVIDIA Corporation in
1999 when the company introduced the first GPU. NVIDIA’s
CUDA architecture (code named Fermi) is the first
architecture to deliver all of the features required for highly
demanding HPC applications. The featuresincludehighlevel
of double-precision floating-point performance, ECC
protection from the registers to DRAM, and support for
languages including C, C++, FORTRAN, Java, MATLAB, and
Python. Fermi is the first complete architecture for GPU
computing [3]. Besides NVIDIA, other GPU vendors include
Intel, ATI, Sony, and IBM.
GPU is specializedforcompute-intensive,highlydata parallel
computation, which what graphics rendering is all about.
Although GPU can be used for 2D data, it is essential for
rendering of 3D animations and video. GPU has a unique
design of ‘many-core’ architecture, and each core is able to
carry out thousands of threads simultaneously.The memory
of GPU consists of a large number of cache blocks, and each
block can be independently accessed [4].
Before GPU was invented, graphics on a personal computer
were performed by a video graphics array (VGA) controller.
The GPU is designed for a class of applications with the
following characteristics [5]:
Computational requirements are large
Parallelism is substantial
Throughput is more important than latency
Parallelism is an integral part of GPU computing and will
become important in future. GPUs offer parallel computing
power that usually requires a computer or a supercomputer
IJTSRD29648](https://image.slidesharecdn.com/137gpucomputinganintroduction-200109113422/85/GPU-Computing-An-Introduction-1-320.jpg)
![International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470
@ IJTSRD | Unique Paper ID – IJTSRD29648 | Volume – 4 | Issue – 1 | November-December 2019 Page 706
to accomplish. Several applications with substantial
parallelism increasingly use the massively parallel
computing capabilities of GPUs to achieve superior
performance [6]. GPUs are leading the way in parallelism;
compilers, algorithms,andcomputational modelshavemade
significant advances in
recent years.
APPLICATIONS
Recently, GPUs are considered not only forgraphics,butalso
for speeding-up theexecutionofgeneral purposealgorithms.
GPU computing has become an increasingly important tool
to develop efficient applications in several areas such as
computer vision, linear algebra, statistical analysis, physics
and biological simulation, database analysis, computational
finance, computational biology, and electronic designs. The
following represent just a few illustrative applications of
GPU computing.
Biological Simulation: Molecular modelling and
molecular design take advantage of scientific
computation capability of GPU. The performance of
GPU-based systems supports the development of
custom-made protocols for efficient modelling of
biomolecular systems and nanostructures.
Computational Electromagnetics: GPUs are used in a
wide range of difference methods or finite-difference
time-domain (FDTD) algorithms, finite element
methods, momentmethods, andMonteCarlomethodsin
science, engineering, computational electromagnetics,
computational fluid dynamics, finance, and acoustics. It
is possible to achieve excellent performance with both
explicit and implicit computational PDEapproximations
on GPUs [7]. The familiar MATLAB package can be used
to take advantage of GPU computing.
Power Systems: In recent years, the computational
demands of modelling modern power systems have
been steadily increasing. With the advent of the smart
grid, the power system has become more complex and
requires more computationally intensive means of
simulation and analysis [8].
Other areas of applications of GPU computing include
medical physics, operations research, financial engineering,
economics, crowd dynamics simulations, optimization, high
performance computing (HPC), neuroscience, atmospheric
modelling,
BENEFTIS AND LIMITATIONS
The key benefit of GPU computing is its massive
performance when compared to the CPU. General purpose
GPU computing has produced the fastest supercomputers in
the world. GPU computing enables applications that we
previously thought impossible due of long execution times.
GPUs have the following advantages [9]:
GPUs are powerful accelerators since they have now
hundreds of computing cores;
GPUs are widely available and relatively cheap;
GPUs require less energy thanothercomputingdevices.
However, GPU computing is only a few years old and the
challenges it is facingare daunting. The major challenges
include memory, arithmetic, and latencies. To be
sustainable, GPU computing must address two open issues:
how to increase applications mean time between failures
and how to minimize unnecessaryenergyconsumption[10].
In some cases, performance cannot scale with the
number of cores because an increasingly large portion
of time is spent on moving data rather than performing
arithmetic operations. GPUs are not designed to excel at
multithreaded access to complex data structuressuch as
a hash table [11].
CONCLUSION
We are in the age of parallel-processing and the age of GPU
computing. GPU has attracted a lot of attention and become
pervasive in today’s computing systems due to its highly
parallel and efficient architecture. Using the GPU for
computing has been an inevitable trend in scientific
community. It is evident that GPU computing will be of great
importance in the near future.Itseemsverypromising.More
information about GPU computing can be found in [11, 12].
REFERENCES
[1] P. Alevras and D. Yurchenko, “GPU computing for
accelerating the numerical path Integrationapproach,”
Computers and Structures, vol. 171, 2016, pp. 46–53.
[2] A. R. Brodtkorbi, T. R. Hageni, and M. L. Satra, “GPU
programming strategiesandtrendsinGPUcomputing,”
Journal of Parallel and Distributed Computing, vol. 73,
no. 1, 2012, pp. 4–13.
[3] P. N. Glaskowsky, “NVIDIA’s Fermi: The first complete
GPU computing architecture,”
http://www.nvidia.com/content/PDF/fermi_white_pa
pers/P.Glaskowsky_NVIDIA%27s_Fermi-
The_First_Complete_GPU_Architecture.pdf
[4] H. Hsieh and C. Chu, “Particle swarm optimization
(PSO)-based tool path planning for 5-axis flank milling
accelerated by graphics processing unit (GPU),”
International Journal of Computer Integrated
Manufacturing, vol. 24, no. 7, 2011, pp. 676-687.
[5] J. D. Owens et al., “GPU Computing,” Proceedings of the
IEEE, vol. 96, no. 5, May 2008, pp. 879-899.
[6] J. Nickolls and W. J. Dally, “The GPU computing era,”
IEEE Micro, March/April 2010, pp. 56-69.
[7] M. Giles et al., “GPU implementation of finite difference
solvers,” Proceedings of the Seventh Workshop on High
Performance Computational Finance, 2014, pp.1-8.
[8] D. J. Sooknanan1 and A. Joshi, “GPU computing using
CUDA in the deployment of smartgrids,”Proceedingsof
SAI Computing Conference, London, UK, July 2016,
pp.1260-1266.
[9] V. Boyer and D. E. Baz, “Recent advances on GPU
computing in operations research,” Proceedings of the
IEEE 27th International Symposium on Parallel &
Distributed Processing Workshops and PhD Forum,
2013, pp. 1778-1787.
[10] J. Y. Shi et al., “Sustainable GPU computing at scale,”
Proceedings of IEEE International Conference on
Computational Science and Engineering, 2011,pp.263-
272.
[11] J. Sanders and E. Kandrot, CUDA by Example: An
Introduction to General-Purpose GPU Programming.
Upper Saddle River, NJ: Addison-Wesley, 2011.
[12] W. W. Hwu, GPU Computing Gems. Burlington, MA:
Elsevier, 2011.](https://image.slidesharecdn.com/137gpucomputinganintroduction-200109113422/85/GPU-Computing-An-Introduction-2-320.jpg)
GPU Computing: An Introduction
- 1. International Journal of Trend in Scientific Research and Development (IJTSRD) Volume 4 Issue 1, December 2019 Available Online: www.ijtsrd.com e-ISSN: 2456 – 6470 @ IJTSRD | Unique Paper ID – IJTSRD29648 | Volume – 4 | Issue – 1 | November-December 2019 Page 705 GPU Computing: An Introduction Matthew N. O. Sadiku, Adedamola A. Omotoso, Sarhan M. Musa Roy G. Perry College of Engineering, Prairie View A&M University, Prairie View, Texas ABSTRACT The graphics processing unit (GPU) is a computer chip that performs rapid mathematical calculations. GPU is a ubiquitous device which appears inevery computing systems such PC, laptop, desktop, and workstation. It is a many- core multithreaded multiprocessor that excels at both graphics and non- graphic applications. GPU computing is using a GPU as a co-processor to accelerate CPUs scientific and engineering computing. The paper provides a brief introduction to GPU computing. KEYWORDS:graphicsprocessingunit, GPUcomputing, heterogeneouscomputing, hybrid computing How to cite this paper: Matthew N. O. Sadiku | Adedamola A. Omotoso | Sarhan M. Musa "GPU Computing: An Introduction" Published in International Journal of Trend in Scientific Research and Development(ijtsrd), ISSN: 2456-6470, Volume-4 | Issue-1, December 2019, pp.705-706, URL: www.ijtsrd.com/papers/ijtsrd29648.pdf Copyright © 2019 by author(s) and International Journal ofTrendinScientific Research and Development Journal. This is an Open Access article distributed under the terms of the Creative CommonsAttribution License (CC BY 4.0) (http://creativecommons.org/licenses/by /4.0) INRODUCTION The efficiency of anycomputersimulationsdependsonthree factors [1]: the formulation of the theory describing the process, numerical methods employed, and the hardware capabilities. Graphics Processing Unit (GPU) computing is regarded today as the most powerful computational hardware. GPU is not the only type of accelerator core that has gained interest recently. Other include field programmable gate array (FPGA) and the Cell Broadband Engine (Cell BE), which receive less attention [2]. Originally, GPU was conceived and developed for rendering graphics. However, due to it high performance and low cost, it has become the new standard of non-graphic application such as image processing, image restoration, filtering, interpolation, and reconstruction. It has become an indispensable part of today's computing systems. In recent years, substantial efforts were made to adapt many algorithms for massively-parallel GPU-based systems since the GPU can perform many calculations simultaneously. GPU computing is the application of a GPU to do general purpose scientific and engineering computing. Central Processing Units (CPUs) are task-parallel processors, while GPUs are data-parallel. GPU computing is not replacing CPU computing. Each approach has its own advantages and limitations. A GPU is used along with a CPU to accelerate scientific and engineering applications.Forthisreason,CPU- GPU computing is also known as heterogeneous computing or hybrid computing. GPU BASICS The term GPU was popularized by NVIDIA Corporation in 1999 when the company introduced the first GPU. NVIDIA’s CUDA architecture (code named Fermi) is the first architecture to deliver all of the features required for highly demanding HPC applications. The featuresincludehighlevel of double-precision floating-point performance, ECC protection from the registers to DRAM, and support for languages including C, C++, FORTRAN, Java, MATLAB, and Python. Fermi is the first complete architecture for GPU computing [3]. Besides NVIDIA, other GPU vendors include Intel, ATI, Sony, and IBM. GPU is specializedforcompute-intensive,highlydata parallel computation, which what graphics rendering is all about. Although GPU can be used for 2D data, it is essential for rendering of 3D animations and video. GPU has a unique design of ‘many-core’ architecture, and each core is able to carry out thousands of threads simultaneously.The memory of GPU consists of a large number of cache blocks, and each block can be independently accessed [4]. Before GPU was invented, graphics on a personal computer were performed by a video graphics array (VGA) controller. The GPU is designed for a class of applications with the following characteristics [5]: Computational requirements are large Parallelism is substantial Throughput is more important than latency Parallelism is an integral part of GPU computing and will become important in future. GPUs offer parallel computing power that usually requires a computer or a supercomputer IJTSRD29648
- 2. International Journal of Trend in Scientific Research and Development (IJTSRD) @ www.ijtsrd.com eISSN: 2456-6470 @ IJTSRD | Unique Paper ID – IJTSRD29648 | Volume – 4 | Issue – 1 | November-December 2019 Page 706 to accomplish. Several applications with substantial parallelism increasingly use the massively parallel computing capabilities of GPUs to achieve superior performance [6]. GPUs are leading the way in parallelism; compilers, algorithms,andcomputational modelshavemade significant advances in recent years. APPLICATIONS Recently, GPUs are considered not only forgraphics,butalso for speeding-up theexecutionofgeneral purposealgorithms. GPU computing has become an increasingly important tool to develop efficient applications in several areas such as computer vision, linear algebra, statistical analysis, physics and biological simulation, database analysis, computational finance, computational biology, and electronic designs. The following represent just a few illustrative applications of GPU computing. Biological Simulation: Molecular modelling and molecular design take advantage of scientific computation capability of GPU. The performance of GPU-based systems supports the development of custom-made protocols for efficient modelling of biomolecular systems and nanostructures. Computational Electromagnetics: GPUs are used in a wide range of difference methods or finite-difference time-domain (FDTD) algorithms, finite element methods, momentmethods, andMonteCarlomethodsin science, engineering, computational electromagnetics, computational fluid dynamics, finance, and acoustics. It is possible to achieve excellent performance with both explicit and implicit computational PDEapproximations on GPUs [7]. The familiar MATLAB package can be used to take advantage of GPU computing. Power Systems: In recent years, the computational demands of modelling modern power systems have been steadily increasing. With the advent of the smart grid, the power system has become more complex and requires more computationally intensive means of simulation and analysis [8]. Other areas of applications of GPU computing include medical physics, operations research, financial engineering, economics, crowd dynamics simulations, optimization, high performance computing (HPC), neuroscience, atmospheric modelling, BENEFTIS AND LIMITATIONS The key benefit of GPU computing is its massive performance when compared to the CPU. General purpose GPU computing has produced the fastest supercomputers in the world. GPU computing enables applications that we previously thought impossible due of long execution times. GPUs have the following advantages [9]: GPUs are powerful accelerators since they have now hundreds of computing cores; GPUs are widely available and relatively cheap; GPUs require less energy thanothercomputingdevices. However, GPU computing is only a few years old and the challenges it is facingare daunting. The major challenges include memory, arithmetic, and latencies. To be sustainable, GPU computing must address two open issues: how to increase applications mean time between failures and how to minimize unnecessaryenergyconsumption[10]. In some cases, performance cannot scale with the number of cores because an increasingly large portion of time is spent on moving data rather than performing arithmetic operations. GPUs are not designed to excel at multithreaded access to complex data structuressuch as a hash table [11]. CONCLUSION We are in the age of parallel-processing and the age of GPU computing. GPU has attracted a lot of attention and become pervasive in today’s computing systems due to its highly parallel and efficient architecture. Using the GPU for computing has been an inevitable trend in scientific community. It is evident that GPU computing will be of great importance in the near future.Itseemsverypromising.More information about GPU computing can be found in [11, 12]. REFERENCES [1] P. Alevras and D. Yurchenko, “GPU computing for accelerating the numerical path Integrationapproach,” Computers and Structures, vol. 171, 2016, pp. 46–53. [2] A. R. Brodtkorbi, T. R. Hageni, and M. L. Satra, “GPU programming strategiesandtrendsinGPUcomputing,” Journal of Parallel and Distributed Computing, vol. 73, no. 1, 2012, pp. 4–13. [3] P. N. Glaskowsky, “NVIDIA’s Fermi: The first complete GPU computing architecture,” http://www.nvidia.com/content/PDF/fermi_white_pa pers/P.Glaskowsky_NVIDIA%27s_Fermi- The_First_Complete_GPU_Architecture.pdf [4] H. Hsieh and C. Chu, “Particle swarm optimization (PSO)-based tool path planning for 5-axis flank milling accelerated by graphics processing unit (GPU),” International Journal of Computer Integrated Manufacturing, vol. 24, no. 7, 2011, pp. 676-687. [5] J. D. Owens et al., “GPU Computing,” Proceedings of the IEEE, vol. 96, no. 5, May 2008, pp. 879-899. [6] J. Nickolls and W. J. Dally, “The GPU computing era,” IEEE Micro, March/April 2010, pp. 56-69. [7] M. Giles et al., “GPU implementation of finite difference solvers,” Proceedings of the Seventh Workshop on High Performance Computational Finance, 2014, pp.1-8. [8] D. J. Sooknanan1 and A. Joshi, “GPU computing using CUDA in the deployment of smartgrids,”Proceedingsof SAI Computing Conference, London, UK, July 2016, pp.1260-1266. [9] V. Boyer and D. E. Baz, “Recent advances on GPU computing in operations research,” Proceedings of the IEEE 27th International Symposium on Parallel & Distributed Processing Workshops and PhD Forum, 2013, pp. 1778-1787. [10] J. Y. Shi et al., “Sustainable GPU computing at scale,” Proceedings of IEEE International Conference on Computational Science and Engineering, 2011,pp.263- 272. [11] J. Sanders and E. Kandrot, CUDA by Example: An Introduction to General-Purpose GPU Programming. Upper Saddle River, NJ: Addison-Wesley, 2011. [12] W. W. Hwu, GPU Computing Gems. Burlington, MA: Elsevier, 2011.