3 ms·
Was the NNUE trained on the CPU though? It’s intuitive that with the small size of a chessboard and the ability to use incremental update of evaluations, it wil
by ummonk 6y ago
Was the NNUE trained on the CPU though? It’s intuitive that with the small size of a chessboard and the ability to use incremental update of evaluations, it will be faster to do neural network evaluation on CPU. Training, on the other hand, is typically highly parallelizable, so it would be a big development if we are able to do it faster on CPU.
- Cybiote 6y agoIt looks like training is on CPU too: https://github.com/nodchip/Stockfish/blob/master/src/nnue/evaluate_nnue_learner.cpp https://github.com/nodchip/Stockfish/blob/master/src/nnue/ev....