GPU architecture describes the design of graphics processing units as massively parallel processors built around thousands of simple shader cores, wide high-bandwidth memory interfaces, and dedicated fixed-function units for texturing, rasterisation, ray tracing, and tensor computation.
Semantic Classification
Content
- A GPU groups thousands of arithmetic units into clusters that execute the same instruction across many threads, fed by a deep memory hierarchy and high-bandwidth memory. Dedicated units handle texturing, rasterisation and increasingly ray tracing and tensor operations.
- This design suits the data-parallel workloads of real-time rendering and general parallel computing. It is exposed to developers through the graphics pipeline and through compute APIs such as CUDA.