Graphics processors or graphics cards are about as old as the PC itself. The first graphics processor unit for PC was introduced in 1981 from IBM and it was called CGA (Color Graphics Adapter). It was the first colour graphics card and also the first colour computer display standard for the IBM PC. IBM CGA graphics cards were equipped with 16 KBytes of video memory and supported several graphics and text modes. The highest resolution of any mode is 640x200 and the highest colour depth supported is 4bit (16 colours). The graphics card was very primitive and mostly consisted of a RAMDAC and memory; there were no specific raster or bliting operations, that means all the processing / rasterization had to be done on the CPU and the result copied from main memory into the display memory. A year after CGA was introduced by IBM another company released the Hercules Graphics Card (HGC) which offered monochrome (1bit) graphics but with a resolution up to 720x348.

In 1984 with the advance of the IBM AT, the successor to the IBM PC, a new graphics adapter was introduced: EGA (Enhanced Graphics Adapter). EGA allowed displaying of 16 colours simultaneously up to a resolution of 640x350. It included 16KByte of ROM to extend the BIOS for additional graphics functions, included the Motorola MC6845 video address generator and had 64Kbyte of video memory which could be extended to 256Kbyte with an additional daughter board.

EGA was superseded by VGA that was introduced by IBM in 1987 and was the last graphics card standard that had been introduced by IBM as of today. VGA differed quite significantly from the previous standard since it was the first card that came with hardware smooth scrolling, hardware split screen, and raster operations. It had 256Kbyte of Video memory and supported 16 and 256 colour modes out of a palette of 262144 colour values (6bit RGB). Also VGA was the first graphics card that incorporated everything into a single chip and therefore could be classified as predecessor to all modern GPUs.

With the downfall of IBM as being the dominating and standard setting company in the PC market and the introduction of windowing systems like Microsoft Windows 2.1, 3.0 and 3.11, IBM OS/2 and Microsoft Windows 95, the graphics card and graphics chips sector started to bloom and many different vendors and graphics chips came onto the market that all differed in terms of features, display resolutions, and functionality. However, this led to a nightmare for application programmers supporting them which meant that most programmers stuck to the VGA standard which all cards were still compatible to (until today). To attack this problem the VESA (Video Electronics Standard Association) introduced VBE (VESA BIOS Extensions) which provides a common interface to access compliant video boards at high resolutions and bit depths and provided access to the card's linear frame buffer.
Many of the graphics cards during that that time offered a level of 2D operation mostly bit blitting, double and triple buffering to accelerate the drawing of widgets and windows.

The first graphics chip to introduce 3D acceleration on PC desktop was the Matrox Impression from Matrox Electronic Systems Ltd., an add-on card for the Matrox Millennium aimed for the PC CAD market. While during that time 3D graphics was still dominated by expensive graphics workstations such as those from SGI introducing OpenGL as the first programmable graphics API for 3D and 2D operations. The first real breakthrough of 3D graphics chip on the PC was the Voodoo Graphics Chip from 3dfx Interactive, a company founded by former SGI employees in 1994. Similar to the Matrox Millennium the 3dfx was a PCI add-on card which didn't come with a RAMDAC by itself but instead daisy-chained to a second standard 2D graphics card. The Voodoo heralded a new era in 3D graphics on the PC and marked the start of the end of expensive graphics workstation. 3dfx introduced their own API called Glide which had a different strategy to all other APIs of that time (Direct3D, OpenGL, QuickDraw3D, and Intel 3DR) it did not hide low-level hardware details behind an "abstraction layer" as all the others did but instead implemented nothing more than what the chip was capable of. At that time it gave Glide a significant performance advantage since the overhead of an abstraction layer was quite costly in terms of memory and CPU processing.

In 1998 NVIDIA introduced the RIVA (Real-time Interactive Video and Animation accelerator) TNT (TwinTexel), NVIDIA's 4th graphics chip at that time. The TwinTexel as its name states introduced a two single texturing pipeline and also offered tri-linear texture filtering. Unlike the Voodo graphics chip and its successor it was fully OpenGL 1.1 compatible. It had a 24bit ZBuffer and full 32bit colour framebuffer.

A year after that NVIDIA released the Geforce256 which NVIDIA marketed "the world's first GPU" and the term Graphics Processing Unit was born. It was the first single-chip processor with integrated transform, lighting, triangle setup/clipping and texturing and rendering engines. A professional version named Quadro signalled NVIDIA's entry into the professional 3D workstation market. And with the Geforce256 it had reached a performance level that rivalled that of high-end professional 3D workstations from SGI that were using the infinite reality engine. The Geforce256 was fully OpenGL 1.2 compatible as well as Direct3D7 which introduced hardware texturing and lighting to the API.