Graphics processing unit: Overview
What it does
The job the technology does for people, in a short paragraph
A graphics processing unit does a very large number of simple calculations at the same time. It was built to draw the images of games and screens, and the first single-chip GPU arrived in 1999 [1]. From 2006 GPUs could also run general programs [3], and they are now widely used to train and run AI models: Nvidia earned 115.2 billion US dollars from its data-centre business in its 2025 fiscal year [5].
as of 2026-11-11
How it works
The working principle in plain words, step by step
A picture on a screen is millions of pixels, and working out the colour of one pixel does not depend on its neighbours. A GPU uses that. Instead of a few powerful cores, as in a central processor, it has many small ones; the GeForce 8800 GTX of 2006 had 128 [3]. The same instruction is sent to many cores at once, and each applies it to its own pixel or number. To draw a scene, the GPU first transforms the corners of each triangle into screen positions, then works out lighting, then fills in the pixels [1]. The same pattern, the same arithmetic on large blocks of numbers, is what neural networks need, which is why GPUs suit the training of neural networks [4].
as of 2026-11-11
Key figures
The handful of numbers that size it up, each with its unit, year and source
| Figure | Value | Unit | Year | Source |
|---|---|---|---|---|
| Transistors on the GeForce 256, the first single-chip GPU | 17 | million | 1999 | [1] |
| Unified stream processors on the GeForce 8800 GTX | 128 | cores | 2006 | [3] |
| Nvidia data-centre revenue, fiscal year ended 26 January 2025 | 115.2 | billion USD | 2025 | [5] |
| Nvidia gaming revenue, fiscal year ended January 2025 | 11.4 | billion USD | 2025 | [5] |
| PC graphics add-in boards shipped worldwide, one quarter (Q2) | 11.6 | million units | 2025 | [6] |
as of 2026-11-11. Revenue figures are Nvidia's own, for its fiscal year ended 26 January 2025; the card shipments are Jon Peddie Research's estimate for the world market.
Where it is used
The main places and products the technology is found in
- Graphics cards for PC gaming and design work
- Games consoles and phones, as part of the main chip
- Data-centre servers that train and run AI models
- Supercomputers for science and engineering simulation
- Cars, where GPUs run AI for driver assistance [2]
as of 2026-11-11
Scale today
How much of it there is now: units in use, made each year and who uses it
- Made per year
- 11.6 million PC graphics add-in boards shipped in the second quarter of 2025 alone [6]
- Unit
- graphics cards
- Year
- 2025
- Note
- Counts add-in cards for PCs only; GPUs built into processors, consoles, phones and AI servers are not included.
as of 2026-11-11
Sources
- Nvidia's GeForce 256: The First Fully Integrated GPU Electronic Design (Jon Peddie), 2022
- Is it Time to Rename the GPU? IEEE Computer Society, Chasing Pixels (Jon Peddie), 2018
- GeForce 8800 GTX: Celebrating 20 Years Of Gaming And CUDA NVIDIA GeForce News (Richard Padilla), 2026
- CHM Releases AlexNet Source Code Computer History Museum (Hansen Hsu), 2025
- NVIDIA Announces Financial Results for Fourth Quarter and Fiscal 2025 NVIDIA Newsroom, 2025
- Q2'25 PC graphics add-in board shipments increased 27.0% from last quarter Jon Peddie Research, 2025