What a GPU is
A GPU, or graphics processing unit, is a processor specialised for doing a very large number of similar calculations at the same time. A graphics card is the board that carries a GPU along with its own fast memory, a cooling system and a connection for the screen. Some computers have a GPU built into the main processor chip instead, with no separate card.
The name comes from the original job: working out what every pixel on the screen should show. The same design turned out to suit other kinds of work as well.
Why drawing a picture is a parallel job
A screen contains a grid of pixels, often millions of them. A typical 1920 by 1080 display has 2,073,600 pixels, and a game may need to update that picture dozens of times each second. For each pixel the computer has to decide a colour, taking into account things like surfaces, lighting and shadows.
The key observation is that the calculation for one pixel mostly does not depend on its neighbours. That means the work can be split up and done simultaneously.
A general-purpose central processing unit has a small number of powerful cores that are good at varied, step-by-step work with lots of decisions. A GPU takes the opposite approach: a very large number of simpler cores that run the same kind of calculation on lots of data at once. Neither design is better overall. They are suited to different tasks, and a computer usually uses both.
What happens when a game draws a frame
Roughly, the sequence for each frame goes like this:
- The game program, running on the main processor, decides what is in the scene: where objects and the camera are.
- It sends that information to the GPU.
- The GPU turns the shapes into the grid of pixels, applies textures (image patterns on surfaces) and computes lighting.
- The finished frame is written into the GPU’s memory and sent to the display.
Shown quickly one after another, these frames create smooth motion. How many appear each second is the frame rate, and the display’s refresh rate sets how many your screen can actually show.
Video memory
Graphics cards have their own memory, often called video memory or VRAM, placed close to the GPU so data can be reached quickly. It holds textures, the frames being built and other data the GPU is working on. Higher resolutions and more detailed scenes generally need more of it. If a program asks for more than is available, it must borrow from slower system memory and performance usually drops, or the program may reduce quality settings.
Video and everyday tasks
Games are the best-known use, but GPUs help elsewhere too:
- Video playback. Many GPUs include dedicated circuits for decoding common video formats, which saves the main processor effort and can save battery on a laptop. This is part of what makes high video resolution and HDR content smooth to watch.
- Video editing and effects. Applying filters, colour changes or transitions across many frames is highly parallel.
- 3D design and rendering. Modelling software leans on the GPU to show objects as you rotate them.
- Ordinary desktop use. Even scrolling a web page and animating windows use graphics hardware, usually through the integrated GPU.
Why artificial intelligence uses GPUs
Many modern artificial intelligence systems are built on neural networks. At the heart of these systems is a huge amount of multiplication and addition applied to large grids of numbers. This is again a case of the same simple operation repeated many times over, and so it maps well onto a GPU’s many cores.
This is why graphics chips became important for training and running AI models, even though they were not originally designed for it. Makers have since added features aimed at that kind of work, and some devices also include separate chips specialised for it. For someone using a chat tool or image generator, the practical effect is that much of the work happens on racks of GPUs in data centres rather than on a home computer, though smaller models can run locally if the hardware is capable.
Integrated versus dedicated graphics
| Integrated | Dedicated | |
|---|---|---|
| Where it lives | Inside the main processor chip | On a separate card or chip |
| Memory | Shares the computer’s main memory | Has its own video memory |
| Power and heat | Low | Higher, often needs stronger cooling |
| Typical use | Everyday work, video, light gaming | Demanding games, 3D and video work |
| Cost | Included | Extra |
Integrated graphics are enough for browsing, documents, streaming and some casual games, and they are why many thin laptops can run for a long time on a charge. A dedicated card makes sense when your main activities involve demanding 3D scenes, large video projects or heavy computation. For most people it is not necessary, and buying more than the work needs mostly adds cost, noise and power use. When weighing options, consider your computer hardware needs as a whole rather than the graphics card alone, since a fast GPU can be held back by too little memory or a weak processor.
Heat, power and cooling
Doing so many calculations at once produces heat. Dedicated cards usually have fans or large heatsinks, and they need extra power from the computer’s supply. In laptops, a powerful GPU often means a thicker body, louder fans and shorter battery life under load. If a chip gets too hot it normally slows itself down to protect itself, which is why a game may run well at first and then become less smooth.
What to look at when comparing
Specifications can be confusing, so a few sensible questions help more than any single number:
- What do you actually plan to do: office work, video calls, games, editing?
- What screen resolution will you use? Higher resolutions are generally more demanding.
- How much video memory does the card have for that use?
- Does your computer’s power supply and case support the card?
Reviews that test the specific programs or games you care about are usually more informative than a comparison of raw specification sheets, because real performance depends on the software as well as the chip.