Commercial Release of Ethereal 13.00 (NNUE) for AVX2 Systems

[AlexChess]

Hi!
Could you also test 12.75 official + perfectly compatible MikeB's nn-c4c306d60d95.nnue https://tests.stockfishchess.org/api/nn ... 60d95.nnue (Stockfish testing nets) . With it on my test the old version is incredibly strong! Since 13.00 github is 3 points weaker than 12.75 github, according to SPCC https://www.sp-cc.de/index.htm after 7000 games, it would be interesting to compare performances using NNUE on both versions.



With nn-c4c306d60d95.nnue :

[AndrewGrant]

Hi!
Could you also test 12.75 official + perfectly compatible MikeB's nn-c4c306d60d95.nnue https://tests.stockfishchess.org/api/nn ... 60d95.nnue (Stockfish testing nets) . With it on my test the old version is incredible strong! Since 13.00 github is 3 points weaker than 12.75 github, according to SPCC https://www.sp-cc.de/index.htm after 7000 games, it would be interesting to compare performances using NNUE on both versions.

No need to do this. Its already known that a Stockfish network will massively out perform what I have now.

[AlexChess]

Hi!
Could you also test 12.75 official + perfectly compatible MikeB's nn-c4c306d60d95.nnue https://tests.stockfishchess.org/api/nn ... 60d95.nnue (Stockfish testing nets) . With it on my test the old version is incredible strong! Since 13.00 github is 3 points weaker than 12.75 github, according to SPCC https://www.sp-cc.de/index.htm after 7000 games, it would be interesting to compare performances using NNUE on both versions.

No need to do this. Its already known that a Stockfish network will massively out perform what I have now.

Not always. I'm testing 12.75 also with more recents 21 MB net and often they do not reach the same performance. Zeus.nnue is very close. I'll test also other free nets.

[AlexChess]

Just started a 100 games match (3 mins + 3 secs 10 mN/s on Banksiagui 0.45 mac M1 4 CPUs) between Ethereal 12.75 NNUE nn-c4c306d60d95.nnue and Ethereal 13.00 standard github (ELO 2400 according to SPCC) to extrapolate if it is stronger than Ethereal 13.00 NNUE commercial (rated 2475 ELO on the same list)



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

[CMCanavessi]

Hey Andrew, been doing some FRC (30m+5s) testing and Ethereal 13.07 (non-bmi avx2, non NNUE) has been crashing. As always, ponder on, 4gb hash, this time I added 6-men syzygy and updated windows 10 with latest bells and whistles. You can check on my twitch, the testing is ongoing but I've already played chiron vs ethereal like 5 times and it crashed in all of them. Maybe something to do with the start position? Dunno...

Here's the last crash:

[pgn][Event "FRC Top 20"]
[Site "CCLS"]
[Date "2021.07.05"]
[Round "1"]
[White "Chiron 4"]
[Black "Ethereal 13.07"]
[Result "1-0"]
[TimeControl "1800+5"]
[Time "12:52:28"]
[Board "10"]
[Termination "crash"]
[Variant "chess960"]
[FEN "bqrnnbkr/pppppppp/8/8/8/8/PPPPPPPP/BQRNNBKR w HChc - 0 1"]
[SetUp "1"]

1. c4 {+0.17/25 116090 122914149} c5 {+0.12/29 57748 78020064} 2. b3 {+0.04/25 59 62285273} b6 {-0.01/30 47179 65035112}
3. Bb2 {+0.07/25 50784 104693754} e6 {+0.01/31 56519 77790112} 4. e3 {+0.00/25 77 58927788} Be7 {+0.01/32 51356 69925787}
1-0
[/pgn]

[AndrewGrant]

Hey Andrew, been doing some FRC (30m+5s) testing and Ethereal 13.07 (non-bmi avx2, non NNUE) has been crashing. As always, ponder on, 4gb hash, this time I added 6-men syzygy and updated windows 10 with latest bells and whistles. You can check on my twitch, the testing is ongoing but I've already played chiron vs ethereal like 5 times and it crashed in all of them. Maybe something to do with the start position? Dunno...

Here's the last crash:

[pgn][Event "FRC Top 20"]
[Site "CCLS"]
[Date "2021.07.05"]
[Round "1"]
[White "Chiron 4"]
[Black "Ethereal 13.07"]
[Result "1-0"]
[TimeControl "1800+5"]
[Time "12:52:28"]
[Board "10"]
[Termination "crash"]
[Variant "chess960"]
[FEN "bqrnnbkr/pppppppp/8/8/8/8/PPPPPPPP/BQRNNBKR w HChc - 0 1"]
[SetUp "1"]

1. c4 {+0.17/25 116090 122914149} c5 {+0.12/29 57748 78020064} 2. b3 {+0.04/25 59 62285273} b6 {-0.01/30 47179 65035112}
3. Bb2 {+0.07/25 50784 104693754} e6 {+0.01/31 56519 77790112} 4. e3 {+0.00/25 77 58927788} Be7 {+0.01/32 51356 69925787}
1-0
[/pgn]

Hmm.. odd. Cannot look into it right now, but launching two tests to see if any crashes occur. Both NNUE and non-NNUE. Any chance you send me the UCI logs from one of these games?

http://chess.grantnet.us/test/12274/ & http://chess.grantnet.us/test/12275/ (Selfplay, just seeing if it ever crashes)

Also that PGN confuses me. The final move made is by black (Ethereal). How would that be an Ethereal crash? I would expect the crashing engine to not make a move? It would have to be crashing on the ponder command it recieves after? UCI logs would be needed. Have you seen crashes against an engine other than Chiron?

[CMCanavessi]

Hmm.. odd. Cannot look into it right now, but launching two tests to see if any crashes occur. Both NNUE and non-NNUE. Any chance you send me the UCI logs from one of these games?

http://chess.grantnet.us/test/12274/ & http://chess.grantnet.us/test/12275/ (Selfplay, just seeing if it ever crashes)

Also that PGN confuses me. The final move made is by black (Ethereal). How would that be an Ethereal crash? I would expect the crashing engine to not make a move? It would have to be crashing on the ponder command it recieves after? UCI logs would be needed. Have you seen crashes against an engine other than Chiron?

It just crashed again
Yes, last night I did a pre-test with shorter TC and it was also crashing (Minic 3.08 also). I don't have logs, but will try to get some. For now, I've replaced the avx2 for the plain avx binary, will report back if it crashes again. Ethereal vs Chiron again after current game.

[AndrewGrant]

Hmm.. odd. Cannot look into it right now, but launching two tests to see if any crashes occur. Both NNUE and non-NNUE. Any chance you send me the UCI logs from one of these games?

http://chess.grantnet.us/test/12274/ & http://chess.grantnet.us/test/12275/ (Selfplay, just seeing if it ever crashes)

Also that PGN confuses me. The final move made is by black (Ethereal). How would that be an Ethereal crash? I would expect the crashing engine to not make a move? It would have to be crashing on the ponder command it recieves after? UCI logs would be needed. Have you seen crashes against an engine other than Chiron?

It just crashed again
Yes, last night I did a pre-test with shorter TC and it was also crashing (Minic 3.08 also). I don't have logs, but will try to get some. For now, I've replaced the avx2 for the plain avx binary, will report back if it crashes again. Ethereal vs Chiron again after current game.

Based on a quick few thousand games ... I can promise its a ponder issue. Why do I have pondering in this engine again? Time to rip it out smh.

[CMCanavessi]

Based on a quick few thousand games ... I can promise its a ponder issue. Why do I have pondering in this engine again? Time to rip it out smh.

It's weird. I changed the avx2 binary to avx, and it still crashed. But now it's playing against Vajolet and so far it looks fine (vs Chiron it crashes almost instantly). Maybe it's opponent-dependant?

[CMCanavessi]

Nope, crashed vs. vajolet2 too. Just took more time.