MMOOZZGG.com

Glossary

What is Nvidia's role in AI?

In short

Nvidia is an American company that makes graphics cards and computing systems for training and running neural networks, and it develops the CUDA software platform.

Companies · Nvidia

MMOOZZGG.com editorsChecked: 1 min read

For a child

Imagine you have to color a huge picture made of a million small squares. One artist would take a month, but a thousand artists with small brushes would finish in an evening. A graphics card is like that team: it has many small calculators that work at the same time. This company makes such teams for computers, and neural networks learn on them.

In plain words

The company started with graphics cards for games, and now its chips are widely used for artificial intelligence. A neural network learns and answers through a huge number of similar calculations, and a GPU runs them in parallel, much faster than an ordinary processor. Nvidia sells single GPUs, servers and complete systems for data centers, which it calls AI factories. Programs for these chips are written with the CUDA platform. For example, a local language model in an office usually runs on one or more of these GPUs.

For business

For a company, Nvidia AI matters in two roles. The first is hardware for your own AI. If a model must run inside your own network, you need a server with GPUs, and their memory decides which model fits. The second is the open Nemotron models. Nvidia publishes them with weights and training data under a license that allows commercial use and running on your own infrastructure. In Russia the situation is different: in March 2022 the company stopped deliveries to the country, and in October 2022 it announced the end of all operations there. So check the terms of hardware service support before you buy, and size the server for specific models.

For a specialist

Nvidia was founded in 1993; co-founder Jensen Huang remains president and CEO. The technical basis of its place in AI is the pairing of GPU and CUDA. In the company’s definition, CUDA is an accelerated computing platform and the software layer through which applications use the GPU. The CUDA Toolkit includes a compiler, libraries and developer tools; code is written in C++, Python or Fortran, or teams use ready accelerated libraries and frameworks. For data centers, the company makes systems it calls AI factories. The Nemotron family covers reasoning, vision, speech, search and safety. The models are distributed under the NVIDIA Open Model License, and as NIM microservices they require an NVIDIA AI Enterprise license.

Example

A company decided to run a language model for contract work in-house. Engineers choose a server with GPUs that have enough memory for the chosen model and check that an answer comes in a couple of seconds.

Related terms

Questions and answers

Why do neural networks need Nvidia GPUs? Training and running a neural network come down to a huge number of identical operations on numbers. A GPU performs thousands of such operations at once, and the CUDA platform gives programs easy access to this computing power.
Does Nvidia work in Russia? No. In March 2022, the company stopped deliveries to Russia, and in October 2022 it announced that it was ending all operations in the country.

Sources

Not sure whether your company needs this?

AI solutions map
Contact us