(1) The architecture of an AI system is the "neural network," which is a matrix of numbers and algorithms implemented in software. The neural network loosely emulates the nerve cells in the human nervous system (see
neural network).
(2) The hardware of an AI system that contains the neural network (definition #1) is typically a collection of servers. Thus, the AI neural network is trained and executed in a server network, although single PCs have also been developed for AI applications (see
AI PC).
Neural Network Nodes and Server Network Nodes
The mathematical components in a neural network are called "nodes," and computing units in a server rack that include a GPU, CPU, memory and storage are also called "nodes." Thus, neural nodes are computed in server nodes.
In AI datacenters, servers are tied together via a communications network typically comprising optical fibers (see
optical fiber). Increasingly, CPUs and GPUs within the server are connected via fiber. See
AI datacenter.