Promising engines 3000+ ELO not Stockfish derivatives

[AlexChess]

Winter 0.9.5 by Winter by Jonathan Rosenthal (Switzerland)

https://github.com/rosenthj/Winter

It is the top rated chess engine from Switzerland and has competed at top invite only computer chess events such as the Top Chess Engine Competition and the Chess.com Computer Championships. Its level of play is super human, but below the state of the art reached by the large, distributed and resource intensive open source projects Stockfish and Leela Chess Zero.

Winter has relied on many machine learning algorithms and techniques over the course of its development. This includes certain clustering methods not used in any other chess programs, such as Gaussian Mixture Models and Soft K-Means.

As of Winter 0.6.2, the evaluation function relies on a small neural network for more precise evaluations.

[AlexChess]

Stash v31.0 by another young genius: Morgan Houppin ELO 3173 (France)
https://www.chessprogramming.org/Morgan_Houppin

"Hi there!

Hi there, version 31 is out ! Here's the list of the changes made: https://github.com/mhouppin/stash-bot/b ... 2021-06-26

Stash scored +62 Elo in regression tests, but even though most of the gains came from eval improvements, they seem to be about ~30-35 Elo against other engines. It now has Ponder support and automated tuning (based on Andrew Grant's paper) integrated to the code base.

Good morning to everyone !

Link: https://gitlab.com/mhouppin/stash-bot

Any feedback would be really appreciated !"

[AlexChess]

Rubichess 2.2 NNUE by Andreas Matthies ELO 2453+ (Germany)

https://github.com/Matthies/RubiChess/releases/tag/2.2
https://www.chessprogramming.org/RubiChess

Andreas also helped me, building his strong engine for Windows ARM64

[AlexChess]

CT800 1.43 Win ARM64 / Ryzen 5900 by Rasmus Althoff (Germany) ELO 3000
https://www.ct800.net/

RAS also kindly compiled for me his engine for Windows 11 ARM64!!!
forum3/viewtopic.php?f=2&t=75911&start=550#p904838


With his own hardware, a microcontroller ARM Cortex-M4, has already 2200+ ELO calculating 43 kN/S, with ARM Silicon M1 4 CPUs is 100 times faster, then with powerful hardware like Ryzen 5900 it becomes a tactical monster !!!!
Just testing it against Fitzbo2 x64 3100 ELO and soon I'll release all PGN games of the preliminary test

[matejst]

Great thread, Alex. Most attention is directed at SF and we miss a lot of other interesting engines. I have tried -- or still use several of this engines -- Wasp (in my peculiar view good at first, then "just another engine", eventually interesting again), Winter (waiting for a new version with a better net), Berserk (good, very interesting, with a huge margin to complete and improve the evaluation), Seer (using it daily now for analysi), and thanks to you, have tried Koivisto (looks good so far). SlowChess is on my list too, alternating with Seer most of the time.

I have tried other engines, but some had bugs. I had the impression that RubiChess pruned too much and missed some tactics (I should have a longer look at the new Rubi 2.2), but I liked it, I had some problems with Clover, and did not find time to test thoroughly some other engines.

In general, while they are good for eng-eng matches, they lack useful features. Nonetheless, with a little patience one can use them and, in general, they are quite good and interesting.

[AlexChess]

Great thread, Alex. Most attention is directed at SF and we miss a lot of other interesting engines. I have tried -- or still use several of this engines -- Wasp (in my peculiar view good at first, then "just another engine", eventually interesting again), Winter (waiting for a new version with a better net), Berserk (good, very interesting, with a huge margin to complete and improve the evaluation), Seer (using it daily now for analysi), and thanks to you, have tried Koivisto (looks good so far). SlowChess is on my list too, alternating with Seer most of the time.

I have tried other engines, but some had bugs. I had the impression that RubiChess pruned too much and missed some tactics (I should have a longer look at the new Rubi 2.2), but I liked it, I had some problems with Clover, and did not find time to test thoroughly some other engines.

In general, while they are good for eng-eng matches, they lack useful features. Nonetheless, with a little patience one can use them and, in general, they are quite good and interesting.

Thank you!
While Stockfish has been developed by a lot of skilled programmers in more than 10 years, original engines like Clover, Stash, Marvin, Winter, Minic are the work of single authors with innovative ideas and they can easily challenge, often draw and sometimes win against it and all his derivates. So yes, I'm really grateful to them and they absolutely deserve my attention. It's very interesting to see how they are continuosly improved

[AlexChess]

https://github.com/jhonnold/berserk/releases/tag/4.5.1

Berserk by Jay Honnold 3315 ELO (USA)

[AlexChess]

CT800 1.43 Win ARM64 / Ryzen 5900 by Rasmus Althoff (Germany) ELO 3000
https://www.ct800.net/

RAS also kindly compiled for me his engine for Windows 11 ARM64!!!
forum3/viewtopic.php?f=2&t=75911&start=550#p904838

With his own hardware, a microcontroller ARM Cortex-M4, has already 2200+ ELO calculating 43 kN/S, with ARM Silicon M1 4 CPUs is 100 times faster, then with powerful hardware like Ryzen 5900 it becomes a tactical monster !!!!
Just testing it against Fitzbo2 x64 3100 ELO and soon I'll release all PGN games of the preliminary test

From the project page https://www.ct800.net/ :
"Software
Author of the ARM-port: Rasmus Althoff.
Base program: NG-Play v9.86 (with extensive improvements and bugfixes).
Author of the base program NG-Play v9.86: George Georgopoulos.
Author of the endgame bitbase for king and pawn vs. king: Marcel van Kervinck.
Opening book: 23,500 different plies in 13,300 unique positions. Transpositions are recognised.
Hash tables: 100 kB.
Pondering: none.
Maximum search depth: 22 plies full search plus 10 plies quiescence.
Maximum game length: 250 moves.
Free ROM: The firmware image is 384 kB, out of which 266 kB are used. 113 kB are for the opening book data and 24 kB for the endgame bitbase.
Free RAM: 16 kB.
Strength: around 2200 ELO."

Since the search depth limit is 22 plies full search plus 10 plies quiescence,
I'm trying to exploit all the CT800 potential.
With my Silicon M1, using Windows 11 ARM64 under Parallels Desktop, it takes 7 minutes to reach the limit (starting position). With a native macOS M1 build (that is 2-3 times faster) I could use regular 3 mins tournament time control.
But my hardware is 10 times slower than a Ryzen 5850x CPU, so I suggest to interested fans to play this match with their better hardware. 

[Ras]

Since the search depth limit is 22 plies full search plus 10 plies quiescence,
I'm trying to exploit all the CT800 potential.

That's only for the Cortex-M4 microcontroller version which is the main scope of my project - the PC UCI version has 42 plies plus 10 plies quiescence.

[AlexChess]

Since the search depth limit is 22 plies full search plus 10 plies quiescence,
I'm trying to exploit all the CT800 potential.

That's only for the Cortex-M4 microcontroller version which is the main scope of my project - the PC UCI version has 42 plies plus 10 plies quiescence.

Effectively after 30 minutes it reached 33/33 ply ...My hardware is too weak, but I'd really like to know CT800 performance with top CPUs. From Cortex-M4 to M1 speed (100x) with 1000x speed could be a quantum leap, a Deep CT800

Best regards, Alex