The graphics card is often the most expensive component in a computer, but for many people it's also the one they could do without. Understanding what a GPU really does and which workloads put it to the test helps you avoid spending too much, or too little.
What a GPU does
A GPU (Graphics Processing Unit) is a processor designed to perform a huge number of simple operations in parallel. It was born to calculate the pixels of an image, but the same architecture lends itself to many other tasks: video processing, effects in editing and photo software, scientific computing, training and running artificial intelligence models.
Besides compute cores, modern GPUs contain dedicated blocks:
- hardware video encoders and decoders, which handle formats such as H.264, HEVC and AV1 with minimal power;
- specialised units for ray tracing in games;
- units for the operations typical of artificial intelligence.
Integrated or dedicated
Integrated graphics is part of the processor and uses system RAM as video memory. Today it's enough for:
- office work, browsing and video calls;
- playing high-resolution video, thanks to hardware decoders;
- light photo editing and simple video editing;
- light or older games at modest settings.
A dedicated card has its own memory (VRAM), its own cooling and much greater computing power. You need one for modern games, professional video editing with effects, 3D modelling and rendering, local AI workloads, and multiple high-resolution monitors with heavy graphics applications.
An often overlooked detail: with integrated graphics, fitting RAM in dual channel noticeably improves graphics performance, because memory is shared.
VRAM
Video memory holds textures, geometry, intermediate images and, in AI workloads, model parameters. When it runs out, the system has to fall back on RAM, which is much slower for the GPU, and performance collapses with obvious stuttering.
Practical guidelines:
| Use | Indicative VRAM |
|---|---|
| Gaming at 1080p | 8 GB as a baseline, more is better for recent titles |
| Gaming at 1440p and 4K | 12-16 GB or more |
| 4K video editing | 8-12 GB, depending on software and effects |
| Local AI models | depends on model size: the more, the better |
The amount of VRAM is not the only parameter: memory bus width, i.e. available bandwidth, matters too. Two cards with the same VRAM can perform differently.
What to check before buying
- Space in the case: modern cards can be over 30 cm long and take up three slots. Measure the case and check the specs.
- Power supply: enough wattage and the right connectors, such as 8-pin or the 16-pin of the latest cards.
- Video outputs: HDMI and DisplayPort in versions compatible with the monitor, especially for high refresh rates.
- PCIe slot: the card works in an older-generation slot too, but some budget cards use only eight or four lanes and lose more on older platforms.
- Drivers and software: some professional programs or compute libraries work better, or only, with GPUs from a particular manufacturer.
Balancing GPU and processor
A very powerful GPU paired with a weak processor will often leave the card underused, especially at low resolutions, where the processor's work weighs more. Conversely, a high-end processor with a modest card will hold the computer back in all graphics workloads. The goal is a balance suited to the type of use.
Power, heat and noise
High-end graphics cards are among the components that draw the most power and generate the most heat. A case with good airflow is essential. Many cards let you cap maximum power or lower the voltage (undervolting) through software: you often lose a small share of performance in exchange for much lower temperatures and noise.
In my work, for office and software development workstations without particular graphics needs, I almost always recommend starting with integrated graphics: a dedicated card can be added later, if it's really needed.
In short
- Integrated graphics is enough for office work, video and light use; with dual-channel RAM it performs better.
- A dedicated GPU is for modern games, professional video, 3D and local AI.
- Insufficient VRAM causes stuttering more than any other limit.
- Check size, power supply, connectors and outputs before buying.
- Aim for a balance between processor and graphics card based on real use.
