Product

CUDA

CUDA is NVIDIA's parallel computing platform and programming model for using NVIDIA GPUs to perform general-purpose computing.

It was introduced in 2006 and became one of the technologies that helped NVIDIA expand beyond computer graphics.

Last reviewed: September 15, 2026

What CUDA is

A GPU contains many processing units designed to perform large numbers of calculations in parallel.

Originally, GPUs were mainly programmed through graphics interfaces.

CUDA gave software developers a more direct way to use that parallel computing capability for tasks that were not necessarily related to drawing images.

This made GPUs useful for areas such as:

  • scientific simulation;
  • engineering;
  • image processing;
  • machine learning;
  • data analytics;
  • computational biology;
  • financial computing;
  • artificial intelligence.

In simple terms

A CPU is designed to handle a broad variety of computing tasks.

A GPU is particularly effective when a problem can be divided into many similar calculations that run at the same time.

CUDA provides the software tools that let developers write programs for that kind of GPU computing.

Why NVIDIA created CUDA

By the mid-2000s, GPUs had become increasingly programmable.

Researchers were already experimenting with graphics hardware for non-graphics calculations, but using GPUs this way was difficult.

NVIDIA introduced CUDA to make GPU computing more accessible through a general-purpose programming model.

That decision changed the role of NVIDIA's hardware.

Instead of being useful mainly for gaming and graphics, NVIDIA GPUs could increasingly act as general-purpose accelerators.

CUDA and artificial intelligence

Modern neural networks require large numbers of matrix and tensor calculations.

These workloads are highly parallel, which makes GPUs well suited to many stages of AI training and inference.

As machine learning expanded, CUDA became an important software layer connecting AI frameworks and applications to NVIDIA hardware.

This relationship between software and hardware helped create a larger ecosystem around NVIDIA GPUs.

CUDA and NVIDIA hardware

CUDA is not a standalone hardware product.

It is part of the broader NVIDIA computing platform.

It is used with NVIDIA GPU architectures including Blackwell and with systems such as DGX.

The exact hardware changes over time, but CUDA gives developers a relatively consistent programming environment across generations of NVIDIA GPUs.

Why CUDA matters to NVIDIA

CUDA is important because it gives developers a reason to build software specifically around NVIDIA's computing platform.

Once research tools, AI frameworks, scientific software, and enterprise applications use CUDA, hardware and software become more closely connected.

This is one reason NVIDIA is better understood as a computing-platform company rather than simply a graphics-chip company.

CUDA was also one of the most important steps in NVIDIA's transition from graphics to AI computing.

Common questions

Is CUDA a programming language?

No. CUDA is a parallel computing platform and programming model. Developers can use languages and tools that support CUDA to write programs that execute work on NVIDIA GPUs.

Is CUDA only for AI?

No. CUDA existed years before the current AI boom and is used in scientific computing, simulation, engineering, data processing, and other fields.

Who owns CUDA?

CUDA was created by NVIDIA and is part of NVIDIA's software ecosystem.

When was CUDA introduced?

NVIDIA introduced CUDA in 2006.

Related

Sources

  • NVIDIA CUDA Programming Guide
  • NVIDIA Corporate Timeline