Drawing a world
A graphics processor performs many related calculations to turn a scene into an image. Dedicated ray-tracing hardware extends the palette of rendering techniques.
NVIDIA · Turing architecture, 2018 ↗One architecture.
Expanding possibilities.
An illustrated guide to three connected shifts—not a benchmark or a claim that one person invented them.
Simulating a world
CUDA makes GPU computing available to programmers beyond graphics. Suitable problems can be divided across many threads; not every task benefits equally.
NVIDIA · CUDA programming guide ↗Learning from a world
Neural networks learn patterns from training data. GPUs accelerate many of the underlying operations; researchers, datasets and software remain essential.
Krizhevsky, Sutskever & Hinton · 2012 paper ↗Original illustrative diagram. Particle speed and count do not represent measured performance.
The chip is only
part of the story.
People & research
Ideas come from researchers, engineers, universities and open scientific exchange.
Tools & software
Libraries, compilers and developer tools help turn hardware capabilities into useful applications.
Manufacturing & systems
Fabrication, packaging, memory, networking and cooling turn designs into working infrastructure.
THE OTHER SIDE OF SCALEMore capability.
More capability.
More responsibility.
Useful AI is more than computational scale. Energy, access, reliability, privacy and the quality of data shape how a technology serves people. This site explains a historical trajectory, not an investment recommendation or an endorsement of every application.
Read the sources and editorial approach ↗