NNUE

NNUE stands for Efficiently Updatable Neural Network. A neural network is a mathematical model trained from examples to turn input data into a useful output. In chess, NNUE receives a representation of the position and produces a value that helps a chess engine judge which possible continuations look more promising.

Its defining idea is efficient updating. From one searched position to the next, only a small part of the board usually changes. NNUE exploits that limited change so much of the network's internal calculation can be reused instead of recomputing everything from scratch at every node. A node is one position examined during the engine's search. The network is trained from large sets of chess positions and values used as learning targets.

NNUE does not replace search. The network evaluates positions while the engine still generates moves, explores variations, and compares outcomes. It should therefore not be confused with the principal variation, the sequence the engine currently considers best and may display to the user. Nor should a version be judged only by its nodes per second: an NNUE evaluation can require more work per node while still leading to stronger decisions.