GPU

A graphics processing unit (GPU) is a specialized processor designed to accelerate the rendering of images and video and to perform parallel computations. Unlike central processing units (CPU) optimized for sequential tasks, GPUs contain hundreds or thousands of smaller cores that can execute many operations concurrently, making them well suited to data-parallel workloads.

GPUs originated to speed up 3D graphics in computer games and evolved from fixed-function pipelines to programmable shaders in the late 1990s and 2000s. The advent of general-purpose computing on GPUs (GPGPU) in the 2000s expanded their use to scientific simulation, engineering, and artificial intelligence.

Modern GPUs have a hierarchical memory model with dedicated graphics memory (VRAM) and large bandwidth. They execute many threads in parallel, typically using SIMT-style execution. Programming interfaces span graphics APIs such as DirectX, Vulkan, and OpenGL, and compute-centric APIs such as CUDA (NVIDIA), HIP/ROCm (AMD), and OpenCL.

GPUs are used for real-time rendering in games and visualization, offline rendering for film and design, and increasingly for high-performance computing tasks, including ML model training and inference, scientific simulations, and data analytics, where large matrix operations are beneficial.

There are discrete GPUs, which have dedicated video memory, and integrated GPUs, which share system memory and are built into CPUs. Vendors include NVIDIA, AMD, and Intel. Consumer lines include GeForce and Radeon, while professional and data-center offerings include NVIDIA RTX/Quadro, AMD Radeon Pro, and Intel Xe.