CUDA Cores in Video Cards

Speed up graphics-intensive processes with Compute Unified Device Architecture

Developed by Nvidia for graphics processing units (GPUs), Compute Unified Device Architecture (CUDA) is a technology platform that accelerates GPU computation processes. Nvidia CUDA cores are parallel or separate processing units within the GPU, with more cores generally equating to better performance.

Product shot of the GeForce GTX TITAN Z graphics card from Nvidia
Nvidia

With CUDA, researchers and software developers can send C, C++, and Fortran code directly to the GPU without using assembly code. This streamlining takes advantage of parallel computing in which thousands of tasks, or threads, are executed simultaneously.

What Are CUDA Cores?

Nvidia CUDA cores are parallel processors similar to a processor in a computer, which may be a dual or quad-core processor. Nvidia GPUs, however, can have several thousand cores.

When shopping for an Nvidia video card, you may see a reference to the number of CUDA cores contained in a card. Cores are responsible for various tasks related to the speed and power of the GPU.

Since CUDA cores are responsible for dealing with all of the data that moves through a GPU, cores often handle video game graphics in situations where characters and scenery are loading.

CUDA cores are similar to AMD's Stream Processors; they're just named differently. However, you cannot equate a 300 CUDA Nvidia GPU with a 300 Stream Processor AMD GPU.

Applications can be built to take advantage of the increased performance offered by CUDA cores. You can see a list of these applications on Nvidia's GPU Applications page.

Choosing a Video Card With CUDA

A higher number of CUDA cores typically means the video card provides faster performance overall. But the number of CUDA cores is only one of several factors to consider when choosing a video card.

Nvidia offers a range of cards featuring as few as eight CUDA cores to as many as 5,760 CUDA cores in the GeForce GTX TITAN Z.

Graphics cards that have Tesla, Fermi, Kepler, Maxwell, or Pascal architecture support CUDA.