2. Current NVIDIA and ATI/AMD architecture for general-purpose computing

Graphics cards from NVIDIA, starting with the G80 series GPU, and from AMD, starting with the R600 series GPU, have the hardware capacity to perform general-purpose floating-point or integer calculations, with application interfaces specific to general-purpose computing (CUDA for NVIDIA and Brook+/CAL at first, then OpenCL for AMD). This makes it easy to adapt the code to these architectures. In this section, we'll take a look at the hardware characteristics of the latest GPUs, which will be useful for controlling computing performance. We'll be referring to GPUs from NVIDIA's Maxwell GM200 series and AMD's R9 Fury series, which are the cards available (September 2015) on the French market. The descriptions given here correspond to the manufacturers' indications, which give an architecture "model". The exact structure of the boards is undoubtedly not identical to this model.