CUDA

CUDA (Compute Unified Device Architecture) is NVIDIA's proprietary parallel computing platform and API, first released in 2006. It allows developers to write C/C++/Python code that runs on NVIDIA GPUs, exploiting the thousands of parallel cores for computations that would be orders of magnitude slower on a CPU.

CUDA is the foundation on which virtually all GPU-accelerated AI frameworks operate — PyTorch, TensorFlow, JAX and others all use CUDA under the hood. CUDA's dominance creates a significant lock-in to NVIDIA hardware: switching to AMD ROCm or Intel oneAPI requires non-trivial porting effort. For data centre operators, CUDA version compatibility between the driver, toolkit and AI frameworks is a frequent operational headache, particularly when managing multiple tenant workloads or upgrading GPU hardware generations.