M1 Apple Silicon for Chess?

[Guenther]

Well I looked at it too and found a lot of serious problems, so much that I don't want to waste any more time on this.

Depths are wildly fluctuating between games and also sometimes inside the same game (for the same programs).

Just one example for inside the same game
(for comparing between games I need to do some stat/calculations, but for now just trust me ;))

You are probably more experienced in investigating PGNs and I trust you definitely.
I just looked at 2-3 games Bluefish vs. Zahak and they didn't use opening book.
So it is even less surprising that Bluefish results show such a big variance if it is allowed to use opening book in some games and not allowed in others.

Regards, Andreas

Time control also is a mystery and I am too lazy too calculate it now.
The earlier games are at a much slower tc than the later ones, despite the timecontrol tag saying it should be always 60+1,
while the chess.com forum posts still tell something about a tc of

Tournament rules: 300 secs /40 + 300 secs /40 + 1 sec /all

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

Same for Syzgy tablebases, here it is telling sth about usage of Syzygy (but not the number of pieces).
Combusken reported +5 in a game for R+B vs. R though ;-)

[AlexChess]

I had a look at your pgn and found only one major problem that leads to strange results like I described:
"No opening book" in your tournament describtion doesn't only mean that the engines are not allowed to use one. It also means that you play every single game from the starting position.

Some may say that this is real chess and the way to test engines. But we can see that this obviously leads to scores like

5% against 3150Elo engines
50% score against 3600Elo engines

for the same single engine.
Of course this is also sss but I wouldn't be surprised if this holds for many more games all played from starting position.

Regards, Andreas

Thank you Andreas, trust me I don't want to troll anybody. I ONLY LIKE CHESS ENGINES, not flames

It's true... I would like to use an opening book for each engine, but not every programmer release his own, so I was thinking that the solution could be to disable all libraries, because using opening books like GOI lead to self moves up to 50 full moves (!) and so isn't the engine to be tested Could you suggest me a solution? (trusted .BIN book openings for engines tests)

Tablebases: actually I'm using only 5-man on Windows 11 ARM64, because 6-man 150 GB Tablebases are in another disk that often disconnects.

TC:

Blitz 60 secs + 1 sec to avoid loses by time on mac M1 native, 2-3 times faster than Windows (about 2-3 min / all on average)

SuperBlitz: 300 secs x 40 moves + 300 secs x 40 moves + 1 sec to avoid loses by time
(about 10-15 mins / all on average)

PGN header could report not updated data. I'm not a professional tester, only a chess computer fan sharing his tournaments with who could be interested (not publishing true rating lists due to few games played, having a limited system)

Best regards, Alex

[mhib]

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

Same for Syzgy tablebases, here it is telling sth about usage of Syzygy (but not the number of pieces).
Combusken reported +5 in a game for R+B vs. R though

The binary I've provided is without syzygy support.

I had to cross-compile, and Go compiler does not support cross-compiling code that uses C interoperability(what is used for interactions with Fathom library).

[AlexChess]

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

Same for Syzgy tablebases, here it is telling sth about usage of Syzygy (but not the number of pieces).
Combusken reported +5 in a game for R+B vs. R though

The binary I've provided is without syzygy support.

I had to cross-compile, and Go compiler does not support cross-compiling code that uses C interoperability(what is used for interactions with Fathom library).

Thank you for the comment. I will check PGN header data and engines config files.

[AlexChess]

Also my darling Igel 3.0.5 for Silicon M1 (Neon) of my kind friend Voffka (Volodymyr Shcherbyna) scores 200 ELO points less than the Windows 11 ARM64 version (compiled directly by him on request) despite it calculates very fast (2x than WinARM64 build)

I hope that a new version or a new NNUE will be out soon

Regards, Alex

[AlexChess]

I had a look at your pgn and found only one major problem that leads to strange results like I described:
"No opening book" in your tournament describtion doesn't only mean that the engines are not allowed to use one. It also means that you play every single game from the starting position.

Some may say that this is real chess and the way to test engines. But we can see that this obviously leads to scores like

5% against 3150Elo engines
50% score against 3600Elo engines

for the same single engine.
Of course this is also sss but I wouldn't be surprised if this holds for many more games all played from starting position.

Regards, Andreas

Maybe Ethereal 13 isn't a regression but, al least on mac M1 native running with 4 CPUs instead of 1 (forced due to Ethereal 13.xx sse3 crashes), with NNUE disabled isn't and upgrade. Two years ago was one of my preferred engines...

ALL PGN GAMES (Ethereal 13.00 github mac M1 vs 12.75 popcnt no NN, 12.75 NN RED, NN Zeus, NN MikeB c4c3)
https://onedrive.live.com/?cid=B0F79706 ... 69&o=OneUp





Best regards, Alex

[AlexChess]

Thank you Acepoint for the new mac M1 updates!!

Best regards and Happy New Year 2022!

PS: If you like, there are also Swordfish 14.7 and Blue Marlin 14.7 TOP SF derivatives GPLv3 sources by dorz to compile
https://www.mediafire.com/file/lm8z4gmp ... s.zip/file

[AlexChess]

INTERESTING POST BY SOPEL ON MAC RUMORS TO IMPROVE M1 PERFORMANCE BY 80%

https://forums.macrumors.com/threads/st ... 0.2326552/

Best Regards, Alex

[AlexChess]

Compiled my ProteusChess SF v004 also for Mac Silicon M1. It's very fast 6-7 Mn/s at starting position using 8 CPUs (modern build on Windows 11 ARM64 calculates 2 Mn/s with 4 CPUs) . I'm testing against all Acepoint's newly updated engines hoping that he will continue to support Mac. Silicon M1 Pro/Max/Ultra are awesome and M1 is becoming interesting also for chess engines.

Proteus Chess SF v004-M1 download link at signature

[Magnum]

Compiled my ProteusChess SF v004 also for Mac Silicon M1. It's very fast 6-7 Mn/s at starting position using 8 CPUs (modern build on Windows 11 ARM64 calculates 2 Mn/s with 4 CPUs) . I'm testing against all Acepoint's newly updated engines hoping that he will continue to support Mac. Silicon M1 Pro/Max/Ultra are awesome and M1 is becoming interesting also for chess engines.

Proteus Chess SF v004-M1 download link at signature

Thx that means Apple devices becomes much better and better for chess.
It would be nice if some Stockfish developers could do improvements for Stockfish running on Apple hardware.