Slightly hairy Blitz game from nightmare.nl chess server

[MikeB]

I wonder, if one would take a very basic engine, and then add NNUE to it, what the result would be. Will a simple engine also gain 200 ELO (just like it will gain about 80-100 if you give it 4 cores instead of one), or will it gain nothing, because NNUE only works well if your normal evaluation is relatively decent? (I don't know; I haven't looked into NNUE yet. Probably won't for a very long time.)

Early implementation suggest NNUE adds value to any engine — for the most part it replaces the evaluate function and the search function is left intact. Most engines adding Stockfish NNUE appears to gain ~200 Elo, whereas Stockfish itself gained initially about 100 Elo roughly and about 50 Elo roughly since then. Weaker engines will gain more than what SF gain by NNUE - mostly because it is appears to be harder to gain ELo as a n engine improves and that SF evaluate function was quite sophisticated and more advanced than majority of other engines. Since NNUE , Stockfish rating Elo growth has accelerated, this can be attributed to restoring interest in SF as more developers are submitting patches and that there was more low hanging fruit available as the change in evaluate function was more accurate now as the result of NNUE. Just my $02, I'm sure there are others who cam articulate their reasons and opinions much better than I can.