Indeed, the graphics processing unit (GPU) has evolved from its original purpose of

accelerating computer graphics and image processing. Initially designed to enhance visual
rendering on video cards and embedded systems like motherboards, mobile phones, and game
consoles, GPUs have found utility beyond graphic-related tasks.
Thanks to their highly parallel structure, GPUs have become instrumental in accelerating
non-graphical computations, particularly for tasks involving embarrassingly parallel
problems. These include complex calculations in fields such as scientific computing, machine
learning, and data analysis. GPUs excel at handling massive amounts of data simultaneously,
making them well-suited for tasks like neural network training and cryptocurrency mining.
The ability of GPUs to perform parallel computations has revolutionized various industries,
enabling researchers, scientists, and developers to tackle computationally intensive tasks
more efficiently. As a result, GPUs have become indispensable components in modern
computing systems, offering significant performance enhancements across a wide range of
applications beyond traditional graphics processing.

Components of a GPU

GPU companies[edit]
Many companies have produced GPUs under a number of brand names. In 2009, [needs
update]
Intel, Nvidia, and AMD/ATI were the market share leaders, with 49.4%, 27.8%, and 20.6%
market share respectively. In addition, Matrox[61] produces GPUs. Modern smartphones use
mostly Adreno GPUs from Qualcomm, PowerVR GPUs from Imagination Technologies, and Mali
GPUs from ARM.
Modern GPUs have evolved to perform a wide range of computational functions beyond
traditional 3D graphics rendering. While they still prioritize graphics-related calculations,
they now include capabilities for basic 2D acceleration and framebuffer operations, often
emulated by 3D hardware in newer cards.
Initially, GPUs were primarily used for memory-intensive tasks like texture mapping and
rendering polygons. Over time, additional units were integrated to accelerate geometric
calculations such as vertex transformations. Recent advancements in GPU technology have
introduced programmable shaders, which allow for complex manipulation of vertices and
textures, as well as techniques like oversampling and interpolation to reduce aliasing.
Several factors influence GPU performance in real-time rendering, including the size of
connector pathways in the semiconductor fabrication, clock signal frequency, and the number
and size of on-chip memory caches. Additionally, the number of streaming multiprocessors
(SM) for Nvidia GPUs, compute units (CU) for AMD GPUs, or Xe cores for Intel discrete
GPUs play a significant role. These units represent the core on-silicon processor units within
the GPU chip that perform calculations, typically working in parallel. GPU performance is
typically measured in floating-point operations per second (FLOPS), with modern GPUs
delivering performance in teraflops (TFLOPS). However, actual display rates may vary due
to other factors.
GPU accelerated video decoding and encoding[edit]

The ATI HD5470 GPU (above, with

copper heatpipe attached) features UVD 2.1 which enables it to decode AVC and VC-1 video
formats.
Since 1995, most GPUs have supported the YUV color space and hardware overlays, which
are crucial for digital video playback. Additionally, many GPUs manufactured since 2000
have incorporated MPEG primitives such as motion compensation and iDCT to further
enhance video decoding capabilities. This hardware-accelerated video decoding, where parts
of the decoding process and video post-processing are offloaded to the GPU hardware, is
commonly known as "GPU accelerated video decoding" or "GPU assisted video decoding".
Modern graphics cards are capable of decoding high-definition video directly on the card,
reducing the workload on the central processing unit (CPU). Various APIs are available for
GPU accelerated video decoding, including DxVA for Microsoft Windows operating
systems, and VDPAU, VAAPI, XvMC, and XvBA for Linux-based and UNIX-like operating
systems. These APIs support decoding videos encoded with a range of codecs such as
MPEG-1, MPEG-2, MPEG-4 ASP, MPEG-4 AVC (H.264), VC-1, WMV3/WMV9,
Xvid/OpenDivX, and DivX 5.
It's worth noting that while most APIs support a wide range of codecs, XvMC is limited to
decoding MPEG-1 and MPEG-2 videos. Additionally, there are dedicated hardware video
decoding and encoding solutions available to further optimize video processing tasks.
Video decoding processes that can be accelerated[edit]
Video decoding processes that can be accelerated by modern GPU hardware are:
 Motion compensation (mocomp)
 Inverse discrete cosine transform (iDCT)
 Inverse telecine 3:2 and 2:2 pull-down correction
 Inverse modified discrete cosine transform (iMDCT)
  In-loop deblocking filter
 Intra-frame prediction
 Inverse quantization (IQ)
 Variable-length decoding (VLD), more commonly known as slice-level acceleration
 Spatial-temporal deinterlacing and automatic interlace/progressive source detection
 Bitstream processing (Context-adaptive variable-length coding/Context-adaptive
binary arithmetic coding) and perfect pixel positioning
These operations also have applications in video editing, encoding, and transcoding.
2D graphics APIs[edit]
An earlier GPU may support one or more 2D graphics API for 2D acceleration, such
as GDI and DirectDraw.[63]
3D graphics APIs[edit]
A GPU can support one or more 3D graphics API, such as DirectX, Metal, OpenGL, OpenGL
ES, Vulkan.

GPU forms[edit]
The term "GPU" has undergone a fascinating evolution, mirroring the rapid advancement of graphics
technology. Initially standing for "graphics processor unit," it denoted a dedicated processing unit
focused solely on graphics tasks. Sony's adoption of the term in 1994, in reference to the
PlayStation's GPU, marked its early usage in the consumer electronics space. However, it was Nvidia's
pivotal marketing move in 1999 that truly popularized the term. With the launch of the GeForce 256,
Nvidia boldly dubbed it "the world's first GPU," emphasizing its transformative integration of various
graphics functions into a single chip. This marked a significant departure from the traditional
approach of separate components for graphics tasks. In response to Nvidia's innovation, ATI
Technologies introduced the term "visual processing unit" (VPU) in 2002 with the Radeon 9700. This
shift in terminology underscored the expanding role of these chips beyond conventional graphics
processing, acknowledging their broader computational capabilities. Fast forward to 2023, and the
landscape continues to evolve with AMD's Alveo MA35D boasting dual VPUs, a testament to ongoing
advancements in performance and efficiency in graphics processing technology.In personal
computers, there are two main forms of GPUs. Each has many synonyms:[70]
 Dedicated graphics also called discrete graphics.
 Integrated graphics also called shared graphics solutions, integrated graphics
processors (IGP), or unified memory architecture (UMA).
Usage-specific GPU[edit]
Most GPUs are designed for a specific use, real-time 3D graphics, or other mass calculations:
1. Gaming
 GeForce GTX, RTX
 Nvidia Titan
 Radeon HD, R5, R7, R9, RX, Vega and Navi series
 Radeon VII
 Intel Arc
2. Cloud Gaming
 Nvidia GRID
 Radeon Sky
3. Workstation
 Nvidia Quadro
 Nvidia RTX
 AMD FirePro
 AMD Radeon Pro
  Intel Arc Pro
4. Cloud Workstation
 Nvidia Tesla
 AMD FireStream
5. Artificial Intelligence training and Cloud
 Nvidia Tesla
 AMD Radeon Instinct
6. Automated/Driverless car
7. Nvidia Drive PX
Dedicated graphics processing units (GPUs) represent a critical component in modern
computing, particularly for tasks involving intensive graphics rendering and
processing. These units are specialized to handle complex graphical calculations,
rendering high-resolution images, and accelerating video playback.
Graphics cards housing dedicated GPUs often feature dedicated RAM, specifically
optimized for graphics-related tasks. This dedicated memory ensures that the GPU has
fast and efficient access to the data it needs, without competing with the system's
main memory.
In desktop computers, dedicated GPUs typically connect to the motherboard via
expansion slots like PCI Express (PCIe) or Accelerated Graphics Port (AGP). This
allows for easy installation and upgradeability, enabling users to enhance their
system's graphics capabilities as needed.
For laptops and other portable devices, dedicated GPUs are often integrated into
proprietary slots or interfaces due to space constraints. Despite this, they still offer
significant performance improvements over integrated graphics solutions, making
them essential for gaming, video editing, and other graphics-intensive applications on
the go.
Advanced technologies like SLI (Scalable Link Interface) and CrossFire allow
multiple GPUs to work together, increasing processing power for demanding tasks.
While less common in consumer systems due to cost and compatibility limitations,
these technologies remain crucial in high-performance computing environments,
including AI training, scientific simulations, and professional graphics workstations.
Overall, dedicated GPUs continue to play a vital role in modern computing, enabling
immersive gaming experiences, smooth video playback, and accelerated graphics
processing for a wide range of applications.
Integrated graphics processing unit[edit]
 The position of an integrated
GPU in a northbridge/southbridge system layout

An ASRock motherboard with

integrated graphics, which has HDMI, VGA and DVI-out ports
Integrated graphics processing units (IGP) have become a prevalent feature in modern
computing systems, offering a balance between cost-effectiveness and graphical
performance. These solutions have evolved significantly over time, transitioning from basic
2D acceleration to handling more sophisticated 3D graphics tasks.
One notable aspect of IGPs is their integration into the motherboard or CPU die, eliminating
the need for a separate graphics card. This integration streamlines system design and reduces
the overall cost of hardware components, making IGPs a popular choice for mainstream
computing devices such as laptops, desktops, and even some mobile devices.
While early IGPs struggled with performance limitations and were often deemed inadequate
for demanding 3D applications, advancements in technology have led to significant
improvements. Modern IGPs, found in AMD APUs and Intel CPUs, boast enhanced graphics
capabilities, allowing them to handle a wide range of tasks, including casual gaming,
multimedia playback, and basic photo and video editing.
Despite their improved performance, IGPs still face challenges related to memory bandwidth
and resource sharing with the CPU. The shared system memory architecture means that both
the CPU and GPU compete for access to RAM, which can impact overall system
performance, especially in graphics-intensive applications.
To address these challenges, manufacturers have introduced innovations such as dedicated
sideport memory and Unified Memory Architecture (UMA). These technologies help
optimize memory usage and improve data transfer efficiency between the CPU and GPU,
resulting in smoother performance and enhanced user experiences.
In conclusion, integrated graphics processing units have come a long way since their
inception, offering viable solutions for everyday computing needs while bridging the gap
between affordability and performance. With ongoing advancements in hardware and
software optimization, IGPs continue to play a crucial role in shaping the future of
computing.
Hybrid graphics processing[edit]
Hybrid GPUs compete with integrated graphics in the low-end desktop and notebook markets.
The most common implementations of this are ATI's HyperMemory and Nvidia's TurboCache.
Hybrid graphics cards are somewhat more expensive than integrated graphics, but much less
expensive than dedicated graphics cards. They share memory with the system and have a small
dedicated memory cache, to make up for the high latency of the system RAM. Technologies
within PCI Express make this possible. While these solutions are sometimes advertised as
having as much as 768 MB of RAM, this refers to how much can be shared with the system
memory.
Stream processing and general-purpose GPUs (GPGPU) have emerged as transformative
technologies, leveraging the formidable computational power of graphics processing units for
a wide array of non-graphical tasks. This paradigm shift, known as GPGPU computing,
harnesses the parallel processing capabilities inherent in GPUs to tackle complex
computational problems with unprecedented efficiency and speed.
Originally designed for rendering graphics, modern GPUs boast highly parallel architectures
optimized for processing vast amounts of data simultaneously. By repurposing the shader
pipeline of a GPU for general-purpose computations, GPGPU computing unlocks new
frontiers in performance, often surpassing the capabilities of traditional CPUs by orders of
magnitude.
Leading GPU manufacturers like AMD and Nvidia have embraced GPGPU computing,
forging partnerships with academic institutions and industry leaders to pioneer
groundbreaking applications across diverse domains. Projects such as Folding@home, a
distributed computing initiative for simulating protein folding, showcase the extraordinary
computational prowess of GPUs in accelerating scientific research and discovery.
GPGPUs excel in tasks characterized by data parallelism, where computations can be
executed concurrently on large datasets. From scientific simulations and computational fluid
dynamics to machine learning and artificial intelligence, GPGPU technology powers
advancements across a spectrum of disciplines.
To facilitate GPGPU programming, GPU vendors provide robust APIs such as OpenCL and
CUDA, empowering developers to craft compute kernels that leverage the GPU's stream
processors efficiently. These programming frameworks offer low-level access to the GPU's
hardware resources, enabling optimized utilization of compute units and memory bandwidth.
In addition to scientific computing and research, GPGPU technology finds applications in
evolutionary computation and genetic programming. By exploiting the parallel processing
capabilities of GPUs, evolutionary algorithms can accelerate fitness evaluations and optimize
solutions across diverse problem domains, driving innovation in optimization and machine
learning.
Overall, GPGPU computing stands at the forefront of high-performance computing,
catalyzing breakthroughs in scientific research, computational modeling, data analytics, and
beyond. As GPU architectures evolve and programming paradigms mature, the
transformative potential of GPGPU computing continues to redefine the boundaries of
computational possibility, ushering in a new era of innovation and discovery.

External GPUs (eGPUs) represent a groundbreaking solution for enhancing the graphics
capabilities of laptops, bridging the gap between portability and performance. As laptops
increasingly become primary computing devices for many users, the need for powerful
graphics processing becomes more pronounced. While laptops may boast ample RAM and a
potent CPU, their onboard graphics chips often lack the horsepower required for demanding
tasks such as gaming, video editing, and 3D rendering.
Enter eGPUs, which offer a transformative solution by enabling users to connect an external
graphics processor to their laptop via a variety of external interfaces, most notably PCI
Express and Thunderbolt. These interfaces provide high-speed data transfer rates, facilitating
seamless communication between the laptop and the external GPU. eGPU enclosures,
equipped with their own power supply units (PSUs) to accommodate the substantial power
demands of modern GPUs, serve as the conduit for this enhanced graphics capability.
The adoption of eGPU technology has gained significant momentum, buoyed by official
support from major hardware vendors and operating system platforms. Apple's decision to
embrace eGPUs with macOS High Sierra 10.13.4 marked a pivotal moment, signaling
broader industry recognition of the potential for eGPU solutions. Leading hardware
manufacturers like HP and Razer have introduced Thunderbolt 3 eGPU enclosures, catering
to the growing demand for external graphics solutions among enthusiasts and professionals
alike.
With eGPUs, users can unlock the full potential of their laptops, transforming them into
powerful workstations capable of handling graphics-intensive tasks with ease. Whether
gaming, content creation, or professional applications, eGPUs offer a flexible and scalable
solution for enhancing graphics performance on-the-go. As eGPU technology continues to
evolve and mature, it promises to redefine the landscape of mobile computing, empowering
users to push the boundaries of what's possible with their portable devices.
Graphics processing units (GPUs) have made significant strides in performance over the
years, but this progress has come with a corresponding increase in energy consumption.
Unlike CPU designers who have shifted focus towards improving performance per watt, GPU
designs have continued to prioritize raw power output. As a result, high-performance GPUs
can draw substantial amounts of power, necessitating the implementation of intelligent power
management techniques.
Efforts to assess GPU efficiency often involve measures like the 3DMark2006 score per watt,
which helps identify more energy-efficient GPUs. However, such metrics may not fully
capture efficiency in typical usage scenarios, where the GPU may spend a significant amount
of time performing less demanding tasks.
In the realm of modern GPUs, energy usage serves as a critical constraint on the achievable
computational capabilities. GPU designs are inherently scalable, allowing manufacturers to
integrate multiple chips on the same video card or use multiple video cards in parallel. The
peak performance of any system is fundamentally limited by its power draw and thermal
dissipation capabilities. Consequently, the performance per watt of a GPU design directly
influences the peak performance attainable by systems employing that design.
Given that GPUs are increasingly used for general-purpose computation, performance
metrics traditionally associated with CPUs, such as FLOPS per watt, are also applicable to
GPUs. This underscores the importance of optimizing GPU designs not only for raw
computational power but also for energy efficiency, enabling more sustainable and
environmentally friendly computing solutions.
Sales[edit]
In 2013, 438.3 million GPUs were shipped globally and the forecast for 2014 was 414.2 million.
However, by the third quarter of 2022, shipments of integrated GPUs totaled around 75.5 million
units, down 19% year-over-year.