Introduction

A graphics processing unit, also occasionally called visual processing unit (VPU), is a specialized electronic circuit designed to rapidly manipulate and alter memory to accelerate the creation of images in a frame buffer intended for output to a display. GPUs are used in embedded systems, mobile phones, personal computers, workstations, and game consoles. Modern GPUs are very efficient at manipulating computer graphics, and their highly parallel structure makes them more effective than general-purpose CPUs for algorithms where processing of large blocks of data is done in parallel. In a personal computer, a GPU can be present on a video card, or it can be on the motherboard or—in certain CPUs—on the CPU die.

The term GPU was popularized by Nvidia in 1999, who marketed the GeForce 256 as the world's first GPU, or Graphics Processing Unit, a single-chip processor with integrated transform, lighting, triangle setup/clipping, and rendering engines that are capable of processing a minimum of 10 million polygons per second.

Modern GPUs use most of their transistors to do calculations related to 3D computer graphics. They were initially used to accelerate the memory-intensive work of texture mapping and renderingpolygons, later adding units to accelerate geometric calculations such as the rotation and translation of vertices into different coordinate systems. Recent developments in GPUs include support for programmable shaders which can manipulate vertices and textures with many of the same operations supported by CPUs, oversampling and interpolation techniques to reduce aliasing, and very high-precision color spaces. Because most of these computations involve matrix and vector operations, engineers and scientists have increasingly studied the use of GPUs for non-graphical calculations.

The History of gpu

The first true 3D graphics started with early display controllers, known as video shifters and video address generators. They acted as a pass-through between the main processor and the display. The incoming data stream was converted into serial bitmapped video output such as luminance, color, as well as vertical and horizontal composite sync, which kept the line of pixels in a display generation and synchronized each successive line along with the blanking interval (the time between ending one scan line and starting the next).

A flurry of designs arrived in the latter half of the 1970s, laying the foundation for 3D graphics as we know them.

RCA’s “Pixie” video chip (CDP1861) in 1976, for instance, was capable of outputting a NTSC compatible video signal at 62x128 resolution, or 64x32 for the ill-fated RCA Studio II console.

The video chip was quickly followed a year later by the Television Interface Adapter (TIA) 1A, which was integrated into the Atari 2600 for generating the screen display, sound effects, and reading input controllers. Development of the TIA was led by Jay Miner, who also led the design of the custom chips for the Commodore Amiga computer later on.

In 1978, Motorola unveiled the MC6845 video address generator. This became the basis for the IBM PC’s Monochrome and Color Display Adapter (MDA/CDA) cards of 1981, and provided the same functionality for the Apple II. Motorola added the MC6847 video display generator later the same year, which made its way into a number of first generation personal computers, including the Tandy TRS-80.

A similar solution from Commodore’s MOS Tech subsidiary, the VIC, provided graphics output for 1980-83 vintage Commodore home computers.

In November the following year, LSI’s ANTIC (Alphanumeric Television Interface Controller) and CTIA/GTIA co-processor (Color or Graphics Television Interface Adaptor), debuted in the Atari 400. ANTIC processed 2D display instructions using direct memory access (DMA). Like most video co-processors, it could generate playfield graphics (background, title screens, scoring display), while the CTIA generated colors and moveable objects. Yamaha and Texas Instruments supplied similar IC’s to a variety of early home computer vendors.

The next steps in the graphics evolution were primarily in the professional fields. Intel used their 82720 graphics chip as the basis for the $1000 iSBX 275 Video Graphics Controller Multimode Board. It was capable of displaying eight color data at a resolution of 256x256 (or monochrome at 512x512). Its 32KB of display memory was sufficient to draw lines, arcs, circles, rectangles and character bitmaps. The chip also had provision for zooming, screen partitioning and scrolling.

SGI quickly followed up with their IRIS Graphics for workstations -- a GR1.x graphics board with provision for separate add-in (daughter) boards for color options, geometry, Z-buffer and Overlay/Underlay.

ATI's Color Emulation Card

The advent of color monitors and the lack of a standard among the array of competitors ultimately led to the formation of the Video Electronics Standards Association (VESA), of which ATI was a founding member, along with NEC and six other graphics adapter manufacturers.

In 1987 ATI added the Graphics Solution Plus series to its product line for OEM’s, which used IBM’s PC/XT ISA 8-bit bus for Intel 8086/8088 based IBM PC’s. The chip supported MDA, CGA and EGA graphics modes via dip switches. It was basically a clone of the Plantronics Colorplus board, but with room for 64kb of memory. Paradise Systems’ PEGA1, 1a, and 2a (256kB) released in 1987 were Plantronics clones as well.

The EGA Wonder series 1 to 4 arrived in March for $399, featuring 256KB of DRAM as well as compatibility with CGA, EGA and MDA emulation with up to 640x350 and 16 colors. Extended EGA was available for the series 2,3 and 4.

Filling out the high end was the EGA Wonder 800 with 16-color VGA emulation and 800x600 resolution support, and the VGA Improved Performance (VIP) card, which was basically an EGA Wonder with a digital-to-analog (DAC) added to provide limited VGA compatibility. The latter cost $449 plus $99 for the Compaq expansion module.

ATI was far from being alone riding the wave of consumer appetite for personal computing.

Many new companies and products arrived that year.. Among them were Trident, SiS, Tamerack, Realtek, Oak Technology, LSI’s G-2 Inc., Hualon, Cornerstone Imaging and Winbond -- all formed in 1986-87. Meanwhile, companies such as AMD, Western Digital/Paradise Systems, Intergraph, Cirrus Logic, Texas Instruments, Gemini and Genoa, would produce their first graphics products during this timeframe.

ATI’s Wonder series continued to gain prodigious updates over the next few years.

In 1988, the Small Wonder Graphics Solution with game controller port and composite out options became available (for CGA and MDA emulation), as well as the EGA Wonder 480 and 800+ with Extended EGA and 16-bit VGA support, and also the VGA Wonder and Wonder 16 with added VGA and SVGA support.

A Wonder 16 was equipped with 256KB of memory retailed for $499, while a 512KB variant cost $699.

An updated VGA Wonder/Wonder 16 series arrived in 1989, including the reduced cost VGA Edge 16 (Wonder 1024 series). New features included a bus-Mouse port and support for the VESA Feature Connector. This was a gold-fingered connector similar to a shortened data bus slot connector, and it linked via a ribbon cable to another video controller to bypass a congested data bus.

The Wonder series updates continued to move apace in 1991. The Wonder XL card added VESA 32K color compatibility and a Sierra RAMDAC, which boosted maximum display resolution to 640x480 @ 72Hz or 800x600 @ 60Hz. Prices ranged through $249 (256KB), $349 (512KB), and $399 for the 1MB RAM option. A reduced cost version called the VGA Charger, based on the previous year’s Basic-16, was also made available.

ATI added a variation of the Wonder XL that incorporated a Creative Sound Blaster 1.5 chip on an extended PCB. Known as the VGA Stereo-F/X, it was capable of simulating stereo from Sound Blaster mono files at something approximating FM radio quality.

The Mach series launched with the Mach8 in May of that year. It sold as either a chip or board that allowed, via a programming interface (AI), the offloading of limited 2D drawing operations such as line-draw, color-fill and bitmap combination (Bit BLIT).

Graphics boards such as the ATI VGAWonder GT, offered a 2D + 3D option, combining the Mach8 with the graphics core (28800-2) of the VGA Wonder+ for its 3D duties. The Wonder and Mach8 pushed ATI through the CAD$100 million sales milestone for the year, largely on the back of Windows 3.0’s adoption and the increased 2D workloads that could be employed with it.

OpenGL. S3 Graphics was formed in early 1989 and produced its first 2D accelerator chip and a graphics card eighteen months later, the S3 911 (or 86C911). Key specs for the latter included 1MB of VRAM and 16-bit color support.

The S3 911 was superseded by the 924 that same year -- it was basically a revised 911 with 24-bit color -- and again updated the following year with the 928 which added 32-bit color, and the 801 and 805 accelerators. The 801 used an ISA interface, while the 805 used VLB. Between the 911’s introduction and the advent of the 3D accelerator, the market was flooded with 2D GUI designs based on S3’s original -- notably from Tseng labs, Cirrus Logic, Trident, IIT, ATI’s Mach32 and Matrox’s MAGIC RGB.

In January 1992, Silicon Graphics Inc (SGI) released OpenGL 1.0, a multi-platform vendor agnostic application programming interface (API) for both 2D and 3D graphics.

OpenGL evolved from SGI’s proprietary API, called the IRIS GL (Integrated Raster Imaging System Graphical Library). It was an initiative to keep non-graphical functionality from IRIS, and allow the API to run on non-SGI systems, as rival vendors were starting to loom on the horizon with their own proprietary APIs.

Initially, OpenGL was aimed at the professional UNIX based markets, but with developer-friendly support for extension implementation it was quickly adopted for 3D gaming.

Microsoft was developing a rival API of their own called Direct3D and didn’t exactly break a sweat making sure OpenGL ran as well as it could under the new Windows operating systems.

Things came to a head a few years later when John Carmack of id Software, whose previously released Doom had revolutionised PC gaming, ported Quake to use OpenGL on Windows and openly criticised Direct3D.

Microsoft’s intransigence increased as they denied licensing of OpenGL’s Mini-Client Driver (MCD) on Windows 95, which would allow vendors to choose which features would have access to hardware acceleration. SGI replied by developing the Installable Client Driver (ICD), which not only provided the same ability, but did so even better since MCD covered rasterisation only and ICD added lighting and transform functionality (T&L).

During the rise of OpenGL, which initially gained traction in the workstation arena, Microsoft was busy eyeing the emerging gaming market with designs on their own proprietary API. They acquired RenderMorphics in February 1995, whose Reality Lab API was gaining traction with developers and became the core for Direct3D.

At about the same time, 3dfx’s Brian Hook was writing the Glide API that was to become the dominant API for gaming. This was in part due to Microsoft’s involvement with the Talisman project (a tile based rendering ecosystem), which diluted the resources intended for DirectX.

As D3D became widely available on the back of Windows adoption, proprietary APIs such as S3d (S3), Matrox Simple Interface, Creative Graphics Library, C Interface (ATI), SGL (PowerVR), NVLIB (Nvidia), RRedline (Rendition) and Glide, began to lose favor with developers.

It didn’t help matters that some of these proprietary APIs were allied with board manufacturers under increasing pressure to add to a rapidly expanding feature list. This included higher screen resolutions, increased color depth (from 16-bit to 24 and then 32), and image quality enhancements such as anti-aliasing. All of these features called for increased bandwidth, graphics efficiency and faster product cycles.

Modern GPU

Until the advent of DirectX 10, there was no point in adding undue complexity by enlarging the die area, which increased vertex shader functionality in addition to boosting the floating point precision of pixel shaders from 24-bit to 32-bit to match the requirement for vertex operations. With DX10's arrival, vertex and pixel shaders maintained a large level of common function, so moving to a unified shader arch eliminated a lot of unnecessary duplication of processing blocks. The first GPU to utilize this architecture was Nvidia's iconic G80.

Aided by the new Coverage Sample anti-aliasing (CSAA) algorithm, Nvidia had the satisfaction of seeing its GTX demolish every single and dual-graphics competitor in outright performance. Despite that success, the company dropped three percentage points in discrete graphics market share in the fourth quarter -- points AMD picked up on the strength of OEM (Original equipment manufacturer) contracts.

The remaining components of Nvidia's business strategy concerning the G80 became reality in February and June of 2007. The C-language based CUDA (Compute Unified Device Architecture) platform SDK (Software Development Kit) was released in beta form to enable an ecosystem leveraging the highly parallelized nature of GPUs. Nvidia's PhysX physics engine as well as its distributed computing projects, professional virtualization and OptiX, Nvidia's ray tracing engine, are the more high profile applications using CUDA.

Both Nvidia and ATI (now AMD) had been integrating ever-increasing computing functionality into the graphics pipeline. ATI/AMD would choose to rely upon developers and committees for the OpenCL path, while Nvidia had more immediate plans in mind with CUDA and high performance computing.

CUDA

Compute Unified Device Architecture is a parallel computing platform and programming model created by NVIDIAand implemented by the graphics processing units (GPUs) that they produce. CUDA gives program developers direct access to the virtual instruction set and memory of the parallel computational elements in CUDA GPUs.

Using CUDA, the GPUs can be used for general purpose processing (i.e., not exclusively graphics); this approach is known asGPGPU. Unlike CPUs, however, GPUs have a parallel throughput architecture that emphasizes executing many concurrent threads slowly, rather than executing a single thread very quickly.

The CUDA platform is accessible to software developers through CUDA-accelerated libraries, compiler directives (such asOpenACC), and extensions to industry-standard programming languages, including C, C++ and Fortran. C/C++ programmers use 'CUDA C/C++', compiled with "nvcc", NVIDIA's LLVM-based C/C++ compiler, and Fortran programmers can use 'CUDA Fortran', compiled with the PGI CUDA Fortran compiler from The Portland Group.

In addition to libraries, compiler directives, CUDA C/C++ and CUDA Fortran, the CUDA platform supports other computational interfaces, including the Khronos Group's OpenCL, Microsoft'sDirectCompute, and C++ AMP. Third party wrappers are also available for Python, Perl, Fortran, Java, Ruby, Lua, Haskell, MATLAB, IDL, and native support in Mathematica.

In the computer game industry, GPUs are used not only for graphics rendering but also in game physics calculations (physical effects like debris, smoke, fire, fluids); examples include PhysX andBullet. CUDA has also been used to accelerate non-graphical applications in computational biology, cryptography and other fields by an order of magnitude or more.

CUDA provides both a low level API and a higher level API. The initial CUDA SDK was made public on 15 February 2007, for Microsoft Windows and Linux. Mac OS X support was later added in version 2.0. which supersedes the beta released February 14, 2008. CUDA works with all Nvidia GPUs from the G8x series onwards, including GeForce, Quadro and the Tesla line. CUDA is compatible with most standard operating systems. Nvidia states that programs developed for the G8x series will also work without modification on all future Nvidia video cards, due to binary compatibility.