For an expansion card that contains a graphics processing unit, seeGraphics card.
Components of a GPU
Agraphics processing unit (GPU) is a specializedelectronic circuit initially designed fordigital image processing and to acceleratecomputer graphics, being present either as a discretevideo card or embedded onmotherboards,mobile phones,personal computers,workstations, andgame consoles. After their initial design, GPUs were found to be useful for non-graphic calculations involvingembarrassingly parallel problems due to theirparallel structure. GPUs' ability to perform vast numbers of calculations rapidly has led to their adoption in diverse fields includingartificial intelligence (AI) where they excel at handling data-intensive and computationally demanding tasks. Other non-graphical uses include the training ofneural networks andcryptocurrency mining. With increasing use of GPUs in non-graphical applications, mainly AI, the term GPU is also erroneously used to describe specialized electronic circuits designed purely for AI and other specialized applications.
Arcade system boards have used specialized graphics circuits since the 1970s. In early video game hardware,RAM for frame buffers was expensive, so video chips composited data together as the display was being scanned out on the monitor.[1]
In 1984,Hitachi released ARTC HD63484, the first majorCMOS graphics processor for personal computers. The ARTC could display up to4K resolution when inmonochrome mode. It was used in a number of graphics cards and terminals during the late 1980s.[13] In 1985, theAmiga was released with a custom graphics chip including ablitter for bitmap manipulation, line drawing, and area fill. It also included acoprocessor with its own simple instruction set, that was capable of manipulating graphics hardware registers in sync with the video beam (e.g. for per-scanline palette switches, sprite multiplexing, and hardware windowing), or driving the blitter. In 1986,Texas Instruments released theTMS34010, the first fully programmable graphics processor.[14] It could run general-purpose code, but it had a graphics-oriented instruction set. During 1990–1992, this chip became the basis of theTexas Instruments Graphics Architecture ("TIGA")Windows accelerator cards.
TheIBM 8514 Micro Channel adapter, with memory add-on
In 1987, theIBM 8514 graphics system was released. It was one of the first video cards forIBM PC compatibles to implementfixed-function 2D primitives inelectronic hardware.Sharp'sX68000, released in 1987, used a custom graphics chipset[15] with a 65,536 color palette and hardware support for sprites, scrolling, and multiple playfields.[16] It served as a development machine forCapcom'sCP System arcade board. Fujitsu'sFM Towns computer, released in 1989, had support for a 16,777,216 color palette.[17] In 1988, the first dedicatedpolygonal 3D graphics boards were introduced in arcades with theNamco System 21[18] andTaito Air System.[19]
In 1991,S3 Graphics introduced theS3 86C911, which its designers named after thePorsche 911 as an indication of the performance increase it promised.[21] The 86C911 spawned a variety of imitators: by 1995, all major PC graphics chip makers had added2D acceleration support to their chips.[22] Fixed-functionWindows accelerators surpassed expensive general-purpose graphics coprocessors in Windows performance, and such coprocessors faded from the PC market.
The term "GPU" was coined bySony in reference to the 32-bitSony GPU (designed byToshiba) in thePlayStation video game console, released in 1994.[31]
In the PC world, notable failed attempts for low-cost 3D graphics chips included theS3ViRGE,ATI Rage, andMatroxMystique. These chips were essentially previous-generation 2D accelerators with 3D features bolted on. Many werepin-compatible with the earlier-generation chips for ease of implementation and minimal cost. Initially, 3D graphics were possible only with discrete boards dedicated to accelerating 3D functions (and lacking 2D graphical user interface (GUI) acceleration entirely) such as thePowerVR and the3dfxVoodoo. However, as manufacturing technology continued to progress, video, 2D GUI acceleration, and 3D functionality were all integrated into one chip.Rendition'sVerite chipsets were among the first to do this well. In 1997, Rendition collaborated withHercules and Fujitsu on a "Thriller Conspiracy" project which combined a Fujitsu FXG-1 Pinolite geometry processor with a Vérité V2200 core to create a graphics card with a full T&L engine years before Nvidia'sGeForce 256; This card, designed to reduce the load placed upon the system's CPU, never made it to market.[citation needed] NVIDIARIVA 128 was one of the first consumer-facing GPU integrated 3D processing unit and 2D processing unit on a chip.
OpenGL was introduced in the early '90s by SGI as a professional graphics API, with proprietary hardware support for 3D rasterization. In 1994Microsoft acquired Softimage, the dominant CGI movie production tool used for early CGI movie hits like Jurassic Park, Terminator 2 and Titanic. With that deal came a strategic relationship with SGI and a commercial license of SGI's OpenGL libraries enabling Microsoft to port the API to the Windows NT OS but not to the upcoming release of Windows '95. Although it was little known at the time, SGI had contracted with Microsoft totransition from Unix to the forthcoming Windows NT OS; the deal which was signed in 1995 was not announced publicly until 1998. In the intervening period, Microsoft worked closely with SGI to port OpenGL to Windows NT. In that era OpenGL had no standard driver model for competing hardware accelerators to compete on the basis of support for higher level 3D texturing and lighting functionality. In 1994 Microsoft announced DirectX 1.0 and support for gaming in the forthcoming Windows '95 consumer OS, in '95Microsoft announced the acquisition of UK based Rendermorphics Ltd and the Direct3D driver model for the acceleration of consumer 3D graphics. The Direct3D driver model shipped with DirectX 2.0 in 1996. It included standards and specifications for 3D chip makers to compete to support 3D texture, lighting and Z-buffering. ATI, which was later to be acquired by AMD, began development on the first Direct3D GPU's. Nvidia, quickly pivoted from afailed deal with Sega in 1996 to aggressively embracing support for Direct3D. In this era Microsoft merged their internal Direct3D and OpenGL teams and worked closely with SGI to unify driver standards for both industrial and consumer 3D graphics hardware accelerators. Microsoft ran annual events for 3D chip makers called "Meltdowns" to test their 3D hardware and drivers to work both with Direct3D and OpenGL. It was during this period of strong Microsoft influence over 3D standards that 3D accelerator cards moved beyond being simplerasterizers to become more powerful general purpose processors as support for hardware accelerated texture mapping, lighting, Z-buffering and compute created the modern GPU. During this period the same Microsoft team responsible for Direct3D and OpenGL driver standardization introduced their own Microsoft 3D chip design calledTalisman. Details of this era are documented extensively in the books: "Game of X" v.1 and v.2 by Russel Demaria, "Renegades of the Empire" by Mike Drummond, "Opening the Xbox" by Dean Takahashi and "Masters of Doom" by David Kushner. TheNvidiaGeForce 256 (also known as NV10) was the first consumer-level card with hardware-accelerated T&L. While the OpenGL API provided software support for texture mapping and lighting, the first 3D hardware acceleration for these features arrived with the firstDirect3D accelerated consumer GPU's.
Nvidia was first to produce a chip capable of programmableshading: theGeForce 3. Each pixel could now be processed by a short program that could include additional image textures as inputs, and each geometric vertex could likewise be processed by a short program before it was projected onto the screen. Used in theXbox console, this chip competed with the one in thePlayStation 2, which used a custom vector unit for hardware accelerated vertex processing (commonly referred to as VU0/VU1). The earliest incarnations of shader execution engines used inXbox were not general purpose and could not execute arbitrary pixel code. Vertices and pixels were processed by different units which had their own resources, with pixel shaders having tighter constraints (because they execute at higher frequencies than vertices). Pixel shading engines were actually more akin to a highly customizable function block and did not really "run" a program. Many of these disparities between vertex and pixel shading were not addressed until theUnified Shader Model.
In October 2002, with the introduction of theATIRadeon 9700 (also known as R300), the world's firstDirect3D 9.0 accelerator, pixel and vertex shaders could implementlooping and lengthyfloating point math, and were quickly becoming as flexible as CPUs, yet orders of magnitude faster for image-array operations. Pixel shading is often used forbump mapping, which adds texture to make an object look shiny, dull, rough, or even round or extruded.[32]
With the introduction of the NvidiaGeForce 8 series and new generic stream processing units, GPUs became more generalized computing devices.Parallel GPUs are making computational inroads against the CPU, and a subfield of research, dubbed GPU computing orGPGPU forgeneral purpose computing on GPU, has found applications in fields as diverse asmachine learning,[33]oil exploration, scientificimage processing,linear algebra,[34]statistics,[35]3D reconstruction, andstock options pricing.GPGPU was the precursor to what is now called a compute shader (e.g. CUDA, OpenCL, DirectCompute) and actually abused the hardware to a degree by treating the data passed to algorithms as texture maps and executing algorithms by drawing a triangle or quad with an appropriate pixel shader.[clarification needed] This entails some overheads since units like thescan converter are involved where they are not needed (nor are triangle manipulations even a concern—except to invoke the pixel shader).[clarification needed]
Nvidia'sCUDA platform, first introduced in 2007,[36] was the earliest widely adopted programming model for GPU computing.OpenCL is an open standard defined by theKhronos Group that allows for the development of code for both GPUs and CPUs with an emphasis on portability.[37] OpenCL solutions are supported by Intel, AMD, Nvidia, and ARM, and according to a report in 2011 by Evans Data, OpenCL had become the second most popular HPC tool.[38]
In 2010, Nvidia partnered withAudi to power their cars' dashboards, using theTegra GPU to provide increased functionality to cars' navigation and entertainment systems.[39] Advances in GPU technology in cars helped advanceself-driving technology.[40] AMD'sRadeon HD 6000 series cards were released in 2010, and in 2011 AMD released its 6000M Series discrete GPUs for mobile devices.[41] The Kepler line of graphics cards by Nvidia were released in 2012 and were used in the Nvidia's 600 and 700 series cards. A feature in this GPU microarchitecture included GPU boost, a technology that adjusts the clock-speed of a video card to increase or decrease it according to its power draw.[42] TheKepler microarchitecture was manufactured.
ThePS4 andXbox One were released in 2013; they both use GPUs based onAMD's Radeon HD 7850 and 7790.[43] Nvidia's Kepler line of GPUs was followed by theMaxwell line, manufactured on the same process. Nvidia's 28 nm chips were manufactured byTSMC in Taiwan using the 28 nm process. Compared to the 40 nm technology from the past, this manufacturing process allowed a 20 percent boost in performance while drawing less power.[44][45]Virtual reality headsets have high system requirements; manufacturers recommended the GTX 970 and the R9 290X or better at the time of their release.[46][47] Cards based on thePascal microarchitecture were released in 2016. TheGeForce 10 series of cards are of this generation of graphics cards. They are made using the 16 nm manufacturing process which improves upon previous microarchitectures.[48] Nvidia released one non-consumer card under the newVolta architecture, the Titan V. Changes from the Titan XP, Pascal's high-end card, include an increase in the number of CUDA cores, the addition of tensor cores, andHBM2. Tensor cores are designed for deep learning, while high-bandwidth memory is on-die, stacked, lower-clocked memory that offers an extremely wide memory bus. To emphasize that the Titan V is not a gaming card, Nvidia removed the "GeForce GTX" suffix it adds to consumer gaming cards.
In 2018, Nvidia launched the RTX 20 series GPUs that added ray-tracing cores to GPUs, improving their performance on lighting effects.[49]Polaris 11 andPolaris 10 GPUs from AMD are fabricated by a 14 nm process. Their release resulted in a substantial increase in the performance per watt of AMD video cards.[50] AMD also released the Vega GPU series for the high end market as a competitor to Nvidia's high end Pascal cards, also featuring HBM2 like the Titan V.
In 2019, AMD released the successor to theirGraphics Core Next (GCN) microarchitecture/instruction set. Dubbed RDNA, the first product featuring it was theRadeon RX 5000 series of video cards.[51] The company announced that the successor to the RDNA microarchitecture would be incremental (a "refresh"). AMD unveiled theRadeon RX 6000 series, its RDNA 2 graphics cards with support for hardware-accelerated ray tracing.[52] The product series, launched in late 2020, consisted of the RX 6800, RX 6800 XT, and RX 6900 XT.[53][54] The RX 6700 XT, which is based on Navi 22, was launched in early 2021.[55]
ThePlayStation 5 andXbox Series X and Series S were released in 2020; they both use GPUs based on theRDNA 2 microarchitecture with incremental improvements and different GPU configurations in each system's implementation.[56][57][58]
Intel firstentered the GPU market in the late 1990s, but produced lackluster 3D accelerators compared to the competition at the time. Rather than attempting to compete with the high-end manufacturers Nvidia and ATI/AMD, they began integratingIntel Graphics Technology GPUs into motherboard chipsets, beginning with theIntel 810 for the Pentium III, and later into CPUs. They began with theIntel Atom 'Pineview' laptop processor in 2009, continuing in 2010 with desktop processors in the first generation of theIntel Core line and with contemporary Pentiums and Celerons. This resulted in a large nominal market share, as the majority of computers with an Intel CPU also featured this embedded graphics processor. These generally lagged behind discrete processors in performance. Intel re-entered the discrete GPU market in 2022 with itsArc series, which competed with the then-current GeForce 30 series and Radeon 6000 series cards at competitive prices.[citation needed]
In the 2020s, GPUs have been increasingly used for calculations involvingembarrassingly parallel problems, such as training ofneural networks on enormous datasets that are needed forlarge language models. Specialized processing cores on some modern workstation's GPUs are dedicated fordeep learning since they have significant FLOPS performance increases, using 4×4 matrix multiplication and division, resulting in hardware performance up to 128 TFLOPS in some applications.[59] These tensor cores are expected to appear in consumer cards, as well.[needs update][60]
Many companies have produced GPUs under a number of brand names. In 2009,[needs update]Intel,Nvidia, andAMD/ATI were the market share leaders, with 49.4%, 27.8%, and 20.6% market share respectively. In addition,Matrox[61] produces GPUs. Chinese companies such asJingjia Micro have also produced GPUs for the domestic market although in terms of worldwide sales, they still lag behind market leaders.[62]
Modern GPUs have traditionally used most of theirtransistors to do calculations related to3D computer graphics. In addition to the 3D hardware, today's GPUs include basic 2D acceleration andframebuffer capabilities (usually with a VGA compatibility mode). Newer cards such as AMD/ATI HD5000–HD7000 lack dedicated 2D acceleration; it is emulated by 3D hardware. GPUs were initially used to accelerate the memory-intensive work oftexture mapping andrendering polygons. Later,units[clarification needed] were added to accelerategeometric calculations such as therotation andtranslation ofvertices into differentcoordinate systems. Recent developments in GPUs include support forprogrammable shaders which can manipulate vertices and textures with many of the same operations that are supported byCPUs,oversampling andinterpolation techniques to reducealiasing, and very high-precisioncolor spaces.
Several factors of GPU construction affect the performance of the card for real-time rendering, such as the size of the connector pathways in thesemiconductor device fabrication, theclock signal frequency, and the number and size of various on-chip memorycaches. Performance is also affected by the number of streaming multiprocessors (SM) for NVidia GPUs, or compute units (CU) for AMD GPUs, or Xe cores for Intel discrete GPUs, which describe the number of on-silicon processor core units within the GPU chip that perform the core calculations, typically working in parallel with other SM/CUs on the GPU. GPU performance is typically measured in floating point operations per second (FLOPS); GPUs in the 2010s and 2020s typically deliver performance measured in teraflops (TFLOPS). This is an estimated performance measure, as other factors can affect the actual display rate.[63]
The ATI HD5470 GPU (above, with copperheatpipe attached) featuresUVD 2.1 which enables it to decode AVC and VC-1 video formats.
Most GPUs made since 1995 support theYUVcolor space andhardware overlays, important fordigital video playback, and many GPUs made since 2000 also supportMPEG primitives such asmotion compensation andiDCT. This hardware-accelerated video decoding, in which portions of thevideo decoding process andvideo post-processing are offloaded to the GPU hardware, is commonly referred to as "GPU accelerated video decoding", "GPU assisted video decoding", "GPU hardware accelerated video decoding", or "GPU hardware assisted video decoding".
In the 1970s, the term "GPU" originally stood forgraphics processor unit and described a programmable processing unit working independently from the CPU that was responsible for graphics manipulation and output.[65][66] In 1994,Sony used the term (now standing forgraphics processing unit) in reference to thePlayStation console'sToshiba-designedSony GPU.[31] The term was popularized byNvidia in 1999, who marketed theGeForce 256 as "the world's first GPU".[67] It was presented as a "single-chipprocessor with integratedtransform, lighting, triangle setup/clipping, and rendering engines".[68] RivalATI Technologies coined the term "visual processing unit" orVPU with the release of theRadeon 9700 in 2002.[69] TheAMD Alveo MA35D features dual VPU’s, each using the5 nm process in 2023.[70]
In personal computers, there are two main forms of GPUs. Each has many synonyms:[71]
Dedicated graphics processing units usesRAM that is dedicated to the GPU rather than relying on the computer’s main system memory. This RAM is usually specially selected for the expected serial workload of the graphics card (seeGDDR). Sometimes systems with dedicateddiscrete GPUs were called "DIS" systems as opposed to "UMA" systems (see next section).[72]
Dedicated GPUs are not necessarily removable, nor does it necessarily interface with the motherboard in a standard fashion. The term "dedicated" refers to the fact thatgraphics cards have RAM that is dedicated to the card's use, not to the fact thatmost dedicated GPUs are removable. Dedicated GPUs for portable computers are most commonly interfaced through a non-standard and often proprietary slot due to size and weight constraints. Such ports may still be considered PCIe or AGP in terms of their logical host interface, even if they are not physically interchangeable with their counterparts.
Graphics cards with dedicated GPUs typically interface with themotherboard by means of anexpansion slot such asPCI Express (PCIe) orAccelerated Graphics Port (AGP). They can usually be replaced or upgraded with relative ease, assuming the motherboard is capable of supporting the upgrade. A few graphics cards still usePeripheral Component Interconnect (PCI) slots, but their bandwidth is so limited that they are generally used only when a PCIe or AGP slot is not available.
Technologies such asScan-Line Interleave by 3dfx,SLI andNVLink by Nvidia andCrossFire by AMD allow multiple GPUs to draw images simultaneously for a single screen, increasing the processing power available for graphics. These technologies, however, are increasingly uncommon; most games do not fully use multiple GPUs, as most users cannot afford them.[73][74][75] Multiple GPUs are still used on supercomputers (like inSummit), on workstations to accelerate video (processing multiple videos at once)[76][77][78] and 3D rendering,[79] forVFX,[80]GPGPU workloads and for simulations,[81] and in AI to expedite training, as is the case with Nvidia's lineup of DGX workstations and servers, Tesla GPUs, and Intel's Ponte Vecchio GPUs.
The position of an integrated GPU in a northbridge/southbridge system layoutAnASRock motherboard with integrated graphics, which has HDMI, VGA and DVI-out ports
Integrated graphics processing units (IGPU),integrated graphics,shared graphics solutions,integrated graphics processors (IGP), orunified memory architectures (UMA) use a portion of a computer's system RAM rather than dedicated graphics memory. IGPs can be integrated onto a motherboard as part of itsnorthbridge chipset,[82] or on the samedie (integrated circuit) with the CPU (likeAMD APU orIntel HD Graphics). On certain motherboards,[83] AMD's IGPs can use dedicated sideport memory: a separate fixed block of high performance memory that is dedicated for use by the GPU. As of early 2007[update] computers with integrated graphics account for about 90% of all PC shipments.[84][needs update] They are less costly to implement than dedicated graphics processing, but tend to be less capable. Historically, integrated processing was considered unfit for 3D games or graphically intensive programs but could run less intensive programs such as Adobe Flash. Examples of such IGPs would be offerings from SiS and VIA circa 2004.[85] However, modern integrated graphics processors such asAMD Accelerated Processing Unit andIntel Graphics Technology (HD, UHD, Iris, Iris Pro, Iris Plus, andXe-LP) can handle 2D graphics or low-stress 3D graphics.
Since GPU computations are memory-intensive, integrated processing may compete with the CPU for relatively slow system RAM, as it has minimal or no dedicated video memory. IGPs use system memory with bandwidth up to a current maximum of 128 GB/s, whereas a discrete graphics card may have a bandwidth of more than 1000 GB/s between itsVRAM and GPU core. Thismemory bus bandwidth can limit the performance of the GPU, thoughmulti-channel memory can mitigate this deficiency.[86] Older integrated graphics chipsets lacked hardwaretransform and lighting, but newer ones include it.[87][88]
On systems with "Unified Memory Architecture" (UMA), including modern AMD processors with integrated graphics,[89] modern Intel processors with integrated graphics,[90] Apple processors, the PS5 and Xbox Series (among others), the CPU cores and the GPU block share the same pool of RAM and memory address space. This allows the system to dynamically allocate memory between the CPU cores and the GPU block based on memory needs (without needing a large static split of the RAM) and thanks to zero copy transfers, removes the need for either copying data over abus (computing) between physically separate RAM pools or copying between separate address spaces on a single physical pool of RAM, allowing more efficient transfer of data.
Hybrid GPUs compete with integrated graphics in the low-end desktop and notebook markets. The most common implementations of this are ATI'sHyperMemory and Nvidia'sTurboCache.
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 highlatency 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)
It is common to use ageneral purpose graphics processing unit (GPGPU) as a modified form ofstream processor (or avector processor), runningcompute kernels. This turns the massive computational power of a modern graphics accelerator's shader pipeline into general-purpose computing power. In certain applications requiring massive vector operations, this can yield several orders of magnitude higher performance than a conventional CPU. The two largest discrete (see "Dedicated graphics processing unit" above) GPU designers,AMD andNvidia, are pursuing this approach with an array of applications. Both Nvidia and AMD teamed withStanford University to create a GPU-based client for theFolding@home distributed computing project for protein folding calculations. In certain circumstances, the GPU calculates forty times faster than the CPUs traditionally used by such applications.[91][92]
GPGPUs can be used for many types ofembarrassingly parallel tasks includingray tracing. They are generally suited to high-throughput computations that exhibitdata-parallelism to exploit the wide vector widthSIMD architecture of the GPU.
GPU-based high performance computers play a significant role in large-scale modelling. Three of the ten most powerful supercomputers in the world take advantage of GPU acceleration.[93]
GPUs support API extensions to theC programming language such asOpenCL andOpenMP. Furthermore, each GPU vendor introduced its own API which only works with their cards:AMD APP SDK from AMD, andCUDA from Nvidia. These allow functions calledcompute kernels to run on the GPU's stream processors. This makes it possible for C programs to take advantage of a GPU's ability to operate on large buffers in parallel, while still using the CPU when appropriate. CUDA was the first API to allow CPU-based applications to directly access the resources of a GPU for more general purpose computing without the limitations of using a graphics API.[citation needed]
Since 2005 there has been interest in using the performance offered by GPUs forevolutionary computation in general, and for accelerating thefitness evaluation ingenetic programming in particular. Most approaches compilelinear ortree programs on the host PC and transfer the executable to the GPU to be run. Typically a performance advantage is only obtained by running the single active program simultaneously on many example problems in parallel, using the GPU'sSIMD architecture.[94] However, substantial acceleration can also be obtained by not compiling the programs, and instead transferring them to the GPU, to be interpreted there.[95] Acceleration can then be obtained by either interpreting multiple programs simultaneously, simultaneously running multiple example problems, or combinations of both. A modern GPU can simultaneously interpret hundreds of thousands of very small programs.
An external GPU is a graphics processor located outside of the housing of the computer, similar to a large external hard drive. External graphics processors are sometimes used with laptop computers. Laptops might have a substantial amount of RAM and a sufficiently powerful central processing unit (CPU), but often lack a powerful graphics processor, and instead have a less powerful but more energy-efficient on-board graphics chip. On-board graphics chips are often not powerful enough for playing video games, or for other graphically intensive tasks, such as editing video or 3D animation/rendering.
Therefore, it is desirable to attach a GPU to some external bus of a notebook.PCI Express is the only bus used for this purpose. The port may be, for example, anExpressCard ormPCIe port (PCIe ×1, up to 5 or 2.5 Gbit/s respectively), aThunderbolt 1, 2, or 3 port (PCIe ×4, up to 10, 20, or 40 Gbit/s respectively), aUSB4 port with Thunderbolt compatibility, or anOCuLink port. Those ports are only available on certain notebook systems.[96] eGPU enclosures include their own power supply (PSU), because powerful GPUs can consume hundreds of watts.[97]
Graphics processing units (GPU) have continued to increase in energy usage, while CPUs designers have recently[when?] focused on improving performance per watt. High performance GPUs may draw large amount of power, therefore intelligent techniques are required to manage GPU power consumption. Measures like3DMark2006 score per watt can help identify more efficient GPUs.[98] However that may not adequately incorporate efficiency in typical use, where much time is spent doing less demanding tasks.[99]
With modern GPUs, energy usage is an important constraint on the maximum computational capabilities that can be achieved. GPU designs are usually highly scalable, allowing the manufacturer to put multiple chips on the same video card, or to use multiple video cards that work in parallel. Peak performance of any system is essentially limited by the amount of power it can draw and the amount of heat it can dissipate. Consequently, performance per watt of a GPU design translates directly into peak performance of a system that uses that design.
Since GPUs may also be used for somegeneral purpose computation, sometimes their performance is measured in terms also applied to CPUs, such as FLOPS per watt.
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.[100][needs update][101]
^Raina, Rajat; Madhavan, Anand; Ng, Andrew Y. (2009-06-14). "Large-scale deep unsupervised learning using graphics processors".Proceedings of the 26th Annual International Conference on Machine Learning – ICML '09. Dl.acm.org. pp. 1–8.doi:10.1145/1553374.1553486.ISBN9781605585161.S2CID392458.
^Barron, E. T.; Glorioso, R. M. (September 1973). "A micro controlled peripheral processor".Conference record of the 6th annual workshop on Microprogramming – MICRO 6. pp. 122–128.doi:10.1145/800203.806247.ISBN9781450377836.S2CID36942876.
V. Garcia and E. Debreuve and M. Barlaud.Fast k nearest neighbor search using GPU. In Proceedings of the CVPR Workshop on Computer Vision on GPU, Anchorage, Alaska, USA, June 2008.
