Commercial Release of Ethereal 13.00 (NNUE) for AVX2 Systems

[AndrewGrant]

Can someone conduct a test of Ethereal 13 NNUE against SF Classical. The last Classical version before NNUE was added into code.

Set Classical contempt at zero.

Some suggest they're probably even in strength if not close.

https://abrok.eu/stockfish/?page=14

Bench: 4746616

Ethereal 13.00 NNUE is weaker than Slowchess 2.6, which is weaker than Stockfish 10.

ie: Buyer Beware (?)

Elo is NOT everything, if that was the case people could just download Stockfisk 13

According to JCER, Ethereal 13:00 (no NNUE) is slight better than SlowChess 2.6. Could someone test 13:00 NNUE against 13:00 no NNUE and 12.75 NNUE?

That seems highly unlikely.

[AlexChess]

New idea

Most reliables rating lists since years test engines with 1,2,4 or 8 CPUs http://www.computerchess.org.uk/ccrl/404/

So it insn't unfair to test Ethereal 13.0.0 github 1CPU if I clearly specify it! Also, other rating lists use only 1 CPU for their test: http://www.cegt.net/40_40%20Rating%20Li ... liste.html I have inserted both github standard and Mikeb's NNUE 190f version to my tournament (now between the top 40 engines, to see if ET can score decently against them, and there are differences using an alien SF NNUE or isn't even loaded. However, it is very fast also with only 1 CPU!!



I will update this topic with all results and the complete pgn file.

[AlexChess]

UPDATE:
I have remembered and also verified that Andrew has told me that Ethereal 13.00 standard doesn't enable NNUEs, so I have replaced it with Ethereal 12.75 NNUE nn-190f102a22c3.nnue (default SF) https://tests.stockfishchess.org/api/nn ... a22c3.nnue



Regards, AlexChess

[pohl4711]

I'm interested also on a test Ethereal 13.00 NNUE commercial vs Ethereal 13.00 standard (github) , it hasn't still conducted by the main rating lists...

http://www.cegt.net/40_40%20Rating%20Li ... liste.html
https://www.sp-cc.de/index.htm
https://ccrl.chessdom.com/ccrl/404FRC/r ... _pure.html

Regards, AlexChess
https://www.chess.com/forum/view/game-a ... e-fritz-17

After the testrun of Stockfish 210628 was finished today, I started the testrun of Ethereal 13.07 avx2 (latest non-NNUE, non-commercial GitHub-version). Result probably on Saturday on my website. I expect a result close to Ethereal 12.75. But we will know it exactly very soon.

Stefan (SPCC)

[AlexChess]

I'm interested also on a test Ethereal 13.00 NNUE commercial vs Ethereal 13.00 standard (github) , it hasn't still conducted by the main rating lists...

http://www.cegt.net/40_40%20Rating%20Li ... liste.html
https://www.sp-cc.de/index.htm
https://ccrl.chessdom.com/ccrl/404FRC/r ... _pure.html

Regards, AlexChess
https://www.chess.com/forum/view/game-a ... e-fritz-17

After the testrun of Stockfish 210628 was finished today, I started the testrun of Ethereal 13.07 avx2 (latest non-NNUE, non-commercial GitHub-version). Result probably on Saturday on my website. I expect a result close to Ethereal 12.75. But we will know it exactly very soon.

Stefan (SPCC)

Thank you Stefan, I'm very interested because I like Ethereal since years and it's a "Tantalus torment" https://en.wikipedia.org/wiki/Tantalus since I cannot test it anymore on my Windows 11 ARM64 running on Parallels Desktop Mac mini M1 (Andrew doesn't support emulations) Strange because Ethereal 12.75 is perfectly stable and very fast, so I'm testing it with all latest SF test NNUEs (just discovered that it works fine only with the 21 Mb nets) and 12.75 + MikeB's NNUE nn-c4c306d60d95.nnue is incredibly stronger than the standard github version (if you want to test it, too).

Regards, Alex

https://www.chess.com/forum/view/game-a ... e-fritz-17

[AlexChess]

@Stefan SPCC Ethereal 13.00 standard github is weaker than Xiphos 0.6 on my test.



Note: all ET13 opponents use also 1CPU and it is the faster, calculating 1.2 mN/s

Regards, AlexChess

[Chessqueen]

@Stefan SPCC Ethereal 13.00 standard github is weaker than Xiphos 0.6 on my test.



Note: all ET13 opponents use also 1CPU and it is the faster, calculating 1.2 Mn/s
Regards, AlexChess

Ethereal is trying hard to keep in the first place of TCEC, if you look at it from the bottom up
https://tcec-chess.com/

[sarona]

Apologies if this has been asked already.

Andrew, I have a purchased copy of Ethereal 13.00 NNUE. I compiled the latest Ethereal development build (13.14) from your GitHub. I have added the commercial net inside the same folder as the executable and set the path in the EvalFile dialogue box.

Will the net be used by the compiled Ethereal 13.14 NNUE engine? I am assuming it would.

It seems to be holding its own against Ethereal 13.00 NNUE in a short test.

[AndrewGrant]

Apologies if this has been asked already.

Andrew, I have a purchased copy of Ethereal 13.00 NNUE. I compiled the latest Ethereal development build (13.14) from your GitHub. I have added the commercial net inside the same folder as the executable and set the path in the EvalFile dialogue box.

Will the net be used by the compiled Ethereal 13.14 NNUE engine? I am assuming it would.

It seems to be holding its own against Ethereal 13.00 NNUE in a short test.

What you compiled on github will not be an NNUE version.
Even if you build an NNUE version, the network will not be compatible due to license differences.
So if you are seeing the two as equal, then something is wrong or perhaps more time is needed to see the massive difference.

[sarona]

You are right. Its not even close now after a few hundred games (3 minutes with 2 second increment). The PVs are very different, too.

Thanks, Andrew.