Generative AI

Generative AI covers artificial intelligence systems that produce something that did not exist before: a written answer, a summary, an image, a voice, a piece of music, a video or computer code. You give it an instruction, called a prompt, and it creates a result based on what it learned during training on very large amounts of data.

Behind these tools there are mainly two families of models. Large language models, or LLMs, built on the Transformer architecture, predict text token by token. Diffusion models, used for images and video, start from random noise and refine it step by step. In both cases the result is probable rather than guaranteed: a generative AI can be confidently wrong, which is why it is worth checking its answers or giving it reliable sources.

These models require a lot of computing power and, above all, a lot of memory. To run them on your own machine, the main criterion is the graphics card's VRAM, because it determines the size of the model you can load. A 16 GB GPU already lets you use mid-sized text models or generate images. Heavier models need 32 GB or more, or even several GPUs.

We offer GPUs, rigs and ready-to-use AI workstations so you can use generative AI on your own premises, without sending your data to an outside service.