Hardware acceleration

Hardware acceleration uses specialized processors or circuits to perform selected computational tasks faster or more efficiently than general-purpose CPU software.

Graphics processing unit: A processor designed to perform many computations in parallel, especially for graphics and numerical workloads. GPUs accelerate workloads that can be divided into many simultaneous operations.
Hardware video decoding: Video decompression performed by dedicated circuitry rather than general-purpose CPU instructions. Dedicated decoders reduce power use during video playback.
Central processing unit: A general-purpose processor that executes program instructions and coordinates computer operations. The CPU provides the flexible baseline that an accelerator complements.
Software optimization: Changes to software that improve its execution speed, memory use, or energy efficiency. Software optimization can improve performance without adding specialized hardware.
Application-specific integrated circuit: An integrated circuit designed for a particular application or a narrow class of tasks. Its fixed-purpose circuitry can execute a target task efficiently.
Ray tracing: A rendering technique that simulates light paths to produce images with realistic reflections, shadows, and illumination. GPU ray-tracing units accelerate recurring geometric intersection calculations.
Parallel computing: The use of multiple computational resources to execute parts of a task simultaneously. Parallelism is a main source of speedup in accelerator workloads.
General-purpose computing: Computing with processors designed to run a broad range of programs rather than one task class. General-purpose processors trade peak task efficiency for flexibility.
Field-programmable gate array: A reconfigurable integrated circuit whose logic can be programmed after manufacturing. FPGAs provide task-specific hardware without fixing the design permanently.
Neural processing unit: A processor specialized for neural-network operations such as matrix multiplication and convolution. NPUs accelerate machine-learning workloads on phones and other devices.

Linked from 3 pages