hardware//NVIDIA//CUDA

CUDA is NVIDIA's GPU programming framework: C-like code that runs on thousands of cores in parallel. Every ML training run, every inference call and every matrix multiply goes through it.

CUDA is NVIDIA's GPU programming framework: C-like code that runs on thousands of cores in parallel. Every ML training run, every inference call and every matrix multiply goes through it.

The same word names the GPU's simplest arithmetic lanes, the CUDA cores, a hardware count quoted on spec sheets; the framework is software that schedules work on those lanes and on the tensor cores beside them (GPU architecture).

CUDA is NVIDIA's moat. PyTorch, TensorFlow and JAX all depend on it, and although AMD has ROCm and Intel has oneAPI, the ecosystem lock-in is massive.