VRAM
The memory on a graphics card. It decides how large a model you can run on your own computer, and it is usually the thing that stops you.
VRAM is the memory built into a graphics card, separate from the main memory of your computer. When you run an AI model locally, the whole model has to fit into that memory at once, together with the text it is currently working on. If it does not fit, the model either runs painfully slowly or refuses to start.
That is why local AI discussions are full of numbers like 8, 16 or 24 gigabytes. As a rough guide, a model’s size in billions of parameters, halved or quartered depending on how heavily it has been compressed, gives you the gigabytes you need. It is also why tools aimed at running capable models on your own machine often set a hard floor, such as the 24GB requirement Perplexity set for its local agent on Windows.