US The Short Answer
CUDA (Compute Unified Device Architecture) is the NVIDIA platform and programming model for running code on a GPU instead of only the CPU. It exposes the thousands of parallel cores in the GPU to software, which is what makes GPU acceleration possible. It is the layer that everything else — cuDNN, TensorRT, and your own code — runs on.
Why It Matters for Edge AI
An edge AI computer pairs a CPU with an NVIDIA GPU. CUDA is what lets an application offload heavy, parallel work — image processing, neural-network maths — onto that GPU. Without it the GPU sits idle and you are left with the CPU alone. It is the reason a compact, low-power edge device can run AI workloads a CPU of the same size could not.
-
//
The foundation cuDNN and TensorRT are built on CUDA; it is the base of the stack.
-
//
Your code too Custom GPU kernels and many vision libraries call CUDA directly.
-
//
Across platforms It runs on NVIDIA GPUs from edge modules to data-centre cards.
CUDA in Practice
CUDA underpins the GPU-accelerated platforms we build edge AI systems on. The CUDA version is set by the platform software, so the application has to match it. For the wider vocabulary, see our guide to common terms in AI GPU computing.