(1) For NVIDIA's classical-quantum computer integration, see
CUDA-Q.
(2) For NVIDIA's software libraries, see
CUDA-X.
(3) (
Compute
Unified
Device
Architecture) NVIDIA's GPU processing units and programming platform, which debuted in 2007. Manufacturers such as AMD, Intel and Apple have their own GPU processing units, although some CUDA applications can be adapted. See
GPU.
CUDA and Tensor Cores
Every NVIDIA GPU has from a few hundred to several thousand CUDA processing cores that operate in parallel, and CUDA is used to program them. CUDA cores perform a simple multiply-and-accumulate operation in parallel. Graphics programmers use CUDA cores but generally not the CUDA interface. The DirectX and OpenGL interfaces are commonly used for 3D graphics.
Although CUDA cores were used for AI, CUDA was created for general purpose processing. After NVIDIA added Google's Tensor processing units to its GPUs in 2017, programmers began to switch to Tensor, because it was created for AI. CUDA was created for general-purpose parallel processing. See
Tensor core.
CUDA C/C++ and CUDA Fortran
CUDA operations are typically programmed in C++ and compiled with NVIDIA's CUDA compiler. A CUDA Fortran compiler was developed by the Portland Group (PGI), which was acquired by NVIDIA. See
AI programming.