CUDA

CUDA is NVIDIA’s parallel computing platform and programming model for running general-purpose computations on graphics processing units. It combines programming interfaces, compiler tools, and a runtime for coordinating CPU and GPU work.

CUDA kernel: A function launched on a GPU to execute in parallel across many threads. Kernels contain the device-side computations CUDA schedules across GPU threads.
Graphics processing unit: A processor designed to perform many computations in parallel, especially for graphics and data-parallel workloads. CUDA targets GPUs as parallel processors rather than only graphics devices.
Deep learning: A machine-learning approach that trains multilayer neural networks using data and numerical optimization. CUDA enables GPU computation used by many deep-learning frameworks and workloads.
OpenCL: An open standard for programming heterogeneous processors, including GPUs from multiple vendors. OpenCL offers a cross-vendor alternative to CUDA’s NVIDIA-specific platform.
CUDA thread hierarchy: CUDA’s organization of threads into blocks and blocks into grids for GPU execution. This hierarchy determines how a kernel’s work is partitioned and scheduled.
NVIDIA: A technology company that designs GPUs, develops CUDA, and produces related computing software. NVIDIA develops CUDA and controls the hardware and software platform it supports.
Scientific computing: The use of computation to model, analyze, and solve problems in science and engineering. CUDA accelerates simulations and numerical methods with abundant parallel work.
ROCm: AMD’s open software platform for GPU computing and accelerated workloads. ROCm is a competing GPU software stack centered on AMD hardware.
CUDA memory model: CUDA’s hierarchy of memory spaces, including registers, shared memory, and device and host memory. Choosing among these spaces governs data access speed and sharing between threads.
Parallel computing: A form of computation in which multiple processing elements work on separate tasks or data simultaneously. CUDA is one platform for expressing and executing parallel computations.

Linked from 8 pages