Graphics processing unit

A graphics processing unit (GPU) is a processor designed to execute many computations in parallel, especially those used to render images. GPUs also accelerate workloads such as machine learning and scientific simulation.

Single instruction, multiple data: A parallel computing model in which one instruction operates on multiple data elements at once. GPU execution groups apply the same operation across many data elements.
Rasterization: The process of converting geometric shapes into pixels or samples on a screen. It is the traditional core task GPUs accelerate to produce real-time images.
Parallel computing: The execution of multiple computations at the same time using multiple processing resources. GPUs are specialized hardware for exploiting large-scale parallelism.
Central processing unit: A general-purpose processor designed to execute a broad range of tasks, often with low latency. CPUs favor flexible control and fast individual threads over GPU-scale throughput.
Streaming multiprocessor: A GPU unit that schedules and executes groups of threads on arithmetic and memory hardware. It is a central building block in many NVIDIA GPU architectures.
Ray tracing: A rendering technique that models light transport by tracing paths through a scene. GPUs accelerate its many independent ray and intersection calculations.
Thread: A sequence of program instructions scheduled for execution by a processor. GPU programs divide work into many lightweight threads.
Tensor processing unit: A specialized processor designed to accelerate tensor operations used in machine learning. TPUs target neural-network workloads more narrowly than general-purpose GPUs.
Shader: A program that computes graphics effects or other per-element operations on a GPU. Shaders express much of the work GPUs perform during graphics rendering.
CUDA: NVIDIA's parallel computing platform and programming model for GPUs. It exposes GPU computation beyond graphics through a widely used software ecosystem.

Linked from 43 pages