How to install Stockfish NNUE!

Stockfish NNUE is a massive revolution in traditional chess engine technology, with a combition of traditional search engine on Neural Network evaluation, resulting in instant improvement of +100 elo to Stockfish.

Stockfish NNUE has two components.

1. Download Stockfish engine

2. Download Neural Network, Neural Network must be kept in the eval folder with  the file name *****.nnue

https://abrok.eu/stockfish/

Then install StockfishNNUE.exe file , the same way you install like Stockfish.

BMI2 could be fastest in most recent cpu, just use whatever is fastest in your cpu.

Right click on the engine display output of Arena.

1. Tick- Use NNUE

2. Locate NNUE file

Note, Speed of SFNNUE will be around 65-70% of classical stockfish speed.

Do you have the link where you can get the map to put  the network in and I was searching on the photo were you paste the network, but I didn't found it.

Thanks for your help. I have now this and in the other folder thisI use scid, I did the normal things to add a new engine and I did start it, but it didn't work. I got this message.Do you know what I am missing?

I have downloaded all the things, but how do you then install the StockfishNNUE.exe file?

Install in GUI as normal engine. For example, in Arena, engine management, install new, choose  SFNNUE as UCI engine. 

What does the nueral network do?

What is it (in this case) and what's the purpose?

Oh, please explain like I'm 5...just the basics.

This Stock evaluation table is completely replaced by Neural Network evaluation.

https://hxim.github.io/Stockfish-Evaluation-Guide/

Every traditional engine has two components,  search and static evaluation. ( table above link). Stockfish search is exactly the same though, afaik.

The purpose is simple, Neural Networks evaluations are better than hand crafted evaluation. Latest tests show that SF nnue is + 50 elo stronger than SF in bullet TC testing.

This guy, the authour of  Rebel engine, organized built in SFnnue for download in his website.

http://rebel13.nl/download/stockfish-nnue.html

Thanks for that...

What are the NN evaluations based on?

It is based on evaluation of milllions of training games of Stockfish at certain depth.

For example, they generate 100 million games at depth 24. Evaluations of Stockfish at depth 24 is saved in nn.bin file.  Then update in another training games.

So, Stockfish nnue, Neural Networks,  learned the outcome at depth 24, without a single search at depth 0. If you generate training games at depth 50, NN evaluations are based on depth 50.

Would you think mostly Lc0 with a better grasp of tactical and endgame positions, like Lc0 as the base and Stockfish as the add-on would work? Or does the change of the base engine even matter at all?

What is this

Ok. I have Lc0 tho. I don't need this

Based on algorithm, Lc0 is based on Generalization. Lco will be excellent in Strategy and positional evaluations, that are critical in opening and early middle game.

Endgame requires discrete calculation, precise search is always a key.. Lc0 will never be as good as Stockfish in endgame as she is x1000 times slower than Stockfish.

Tests have been shown that using Lco in first 35 moves and  SF in endgame is +80 elo stronger than either Stockfish. 

How about performance. LC0 is pretty useless for the majority of CPU only users. I am assuming that SFNNUE uses Neural Nets so to run it only on a CPU is equally useless. The debate again comes down to how much you want to invest in a decent GPU (or GPUs) to get the optimum/maximum performance from the NN Engine and the corresponding net, whether SFNNUE or LC0.

The developers really need to concentrate on how to run NNs on CPU only machines. There is no point if the average user has to buy GPUs with tensor cores to run Chess Engines for daily use.

Dude, you need to know what is NNUE first  before talking about nonsense.

https://www.chess.com/forum/view/general/stockfishnnue-probably-the-future-of-stockfish

Many themes of puzzles are used to test various engines. Here is a collection of 213 such puzzles where engines mostly fail as the themes are really complex. I tested SF, SFNNUE and LC0 (all on CPU) and only LC0 managed to solve about 50% of these. SF and SFNNUE failed miserably. It would be great if someone can make an objective comparison and post results. @drmrboss you can keep talking rude nonsense sitting on your couch. The rest of us will get on with adding some value in the World.

#1 to #60 : https://lichess.org/study/y6sM5eAf
#61 to #120 : https://lichess.org/study/Vfe59GOH
#121 to #180 : https://lichess.org/study/0FQa6riN
#181 to #213 : https://lichess.org/study/s2mTErEZ

@drmrboss I am on Discord. Github is for foetuses like you.

Thanks! These are fast moving developments, please be nice people !  This is all very interesting .  Are there NNUE nets trained on deep LC0 evaluation of positions?  This would then allow LC0 evaluations in an AB engine.