Chess Engines binaries for Raspberry Pi 4 Linux 64 bits

[Pi4Chess]

Hello,

This post to share the open source engines i have managed to compile or found for my little Raspberry Pi 4 with Ubuntu Mate 64 bits.

These are working fine under Arena 3.9Beta for Linux aarch64 version (http://www.playwitharena.de/downloads/a ... eta.tar.gz).

For an idea of engines performance in k/ns with this platform, its 4 cores are like 75 to 100% of 1 core of an old i2500K.

The list of engines :

• Cfish 12 (official armv8 compile)
  Stockfish 12
  Crystal 3.0
  CT800 V1.41
  Defenchess 2.2
  Ethereal 12.75
  Halogen 8.1
  Igel 2.8
  K2_096-dev (not using all the cores even if 4 threads in GUI)
  Monolith 2
  Protector
  Rodent II
  Rodent III
  RubiChess 1.8
  Senpai 2 (not using all the cores even if 4 threads in GUI)
  Stash bot 23
  Texel 1.08a

You can download them here (with last SF nnue net as for 24/11/20) :
https://pixeldrain.com/u/H8ZpoLqf


Also you get with Arena 3.9beta for Linux the following engines :

• Arasan 21.4
  K2 v.087
  Stockfish 10
  TogaII 4.0.1

From repositories you can find also Stockfish 11 and old xboard engines like Crafty, Fruit etc...

May be some of them could get little compilation tweaks by a pro compiler. If engine authors could give some love to this platform including the right profile and flags in their makefile it would be awesome.
Also feel free to share your compilations by replying to this thread if you have other engines or if yours are more optimized and have better k/ns performance.

Special demand : if someone could compile the crafty 25.6 uci for this little machine it would be awesome since i couldn't do it and Crafty has always been one of my old loved engine.

Special thanks to Rasmus Althoff for his initial help.

Have fun

[Pi4Chess]

Compiled 4 more engines :

-Mr Bob chess 0.9
-Vajolet 2.8
-Nemorino 6
-Weiss 1.2

Added a second file for them ; https://pixeldrain.com/u/jrsJtpLo

[George Sobala]

lc0 will also compile

Code: Select all

git clone https://github.com/LeelaChessZero/lc0.git
cd lc0
git submodule update --init --recursive
pip3 install meson
sudo apt install cmake
pip3 install ninja
./build.sh -Dgtest=false

[Pi4Chess]

lc0 will also compile

Code: Select all

git clone https://github.com/LeelaChessZero/lc0.git
cd lc0
git submodule update --init --recursive
pip3 install meson
sudo apt install cmake
pip3 install ninja
./build.sh -Dgtest=false

Thank you for this input I confess I did not even try since this engine is more GPU efficient but will give it a try.

[Pi4Chess]

7 more engines !

1 found on Github :

• -ArmFish : the ARM compiled version of AsmFish (last date 27-03-2018)/
  Nice 2500kn per second but no nnue :-/

6 compiled :

• -lc0 0.26.3 compiled thx to George Sobala. The "performance" is about 6 to 7 depth per move for a 10 min game with a 10b x128 f net (included in file).
  -Minic 3
  -BrainLearn 12.1
  -Demolito 09/20
  -ShashChess 15.1
  -Marvin 4

Binaries can be dowloaded there : https://pixeldrain.com/u/rfYNF6kk

A total of 34 chess engines available (included SF 11 and crafty from repositories)

[Pi4Chess]

Updated engines :

-Stash Bot v24
-Rubichess 1.9 NN with depth 16 net (engines seems to be stronger without NN at the moment of post with very little testing)

Download link : https://pixeldrain.com/u/4xH3DDW9

[Pi4Chess]

Added a bunch of compiled engines (many oldies winboard and many that only use 1 core of cpu) here is the list :

-Rodent IV (version that can play 960 too, lagging threads)
-Chess-at-nite
-Feeks
-Fruit 2.1
-Leonidas r83
-WyldChess
-OpenTal 1.0 (more agressive than 1.1)
-OpenTal 1.1
-Sunfish
-Betsabe II
-Bismarck
-Chesley
-Claudia
-Diablo051
-Discocheck
-Faile1.4
-Floyd 0.9
-Galjoen 0.40.1
-Gambit Fruit (gfruit)
-GreKo 2020-03
-Gully 2
-Jazz r840
-Queen Maxima 2
-MORA
-Phalanx XXV
-Prophet 3
-Prophet 4 (lagging threads)
-PurpleHaze 2.1
-QBBEngine
-Sayuri
-Sjaak II 1.4.1
-Stupid
-Sungorus 1.4
-Toledo Nano (2kbytes Winboard engine !)
-Viper 01
-Zetadva
-Zevra 2
-Zevra 1.8 (seems better than zevra 2!)

Download file there : https://pixeldrain.com/u/dvZCVXiH

That's a total of 72 engines to toy with on Raspberry Pi 4 running Linux 64 bits Ubuntu (should work on other 64 bits distros).

[Pi4Chess]

Here are 9 more engines i managed to compile for the little raspberry Pi 4. Some are new ones or at least active ones that may have some updates in the future.

This sums to 81 available engines posted in this topic !

-Beef 0.3.6
-Chessika 2.27
-Drofa 2
-Dumb 1.5
-FracTal
-Payfleens
-Pulse 1.7.2
-Sapeli 2 Final
-Supernova 2.21

Download link : https://pixeldrain.com/u/vi48Wj1u

I think the updates will be less frequent now since there are still other engines out there but i did not managed to compile them for various reasons (no makefile, errors at compiling that i could not resolve, windows oriented source etc...).
Feel free to help the cause if you managed to compile a chess engine that is not in this list or if your compiled version has more performance (more kns, using all the cores etc...)

Hace fun !

[mvanthoor]

I still wonder if these engines keep working if the kernel and/or glibc are updated on the Raspberry Pi.

To be honest, I don't know how I would create Linux or Mac executables of my engine.

If I create the executable using the latest Debian, will it run under the latest Fedora? If I create the executable under the latest Raspberry Pi OS on the RPi 4, will it run under a newer version of the same operating system? Same for the Mac: if I create it to run under Catalina, will it still run under Big Sur (and there needs to be yet another executable for the M1 CPU.) Android: there are so many Android phones and versions around that I can't even begin to fathom what I would need.

The only thing I know for sure is that if I create an executable under Windows, it will work on any Windows version, if the following conditions are true:
- The correct version of the Visual Studio C++ redistributable is installed (most systems have every single one of them installed since 2005, and they are maintained by Windows Update)
- The CPU actually supports the executable.

For Linux, Mac and Android, I don't know... I don't have the systems or the time to create and test 50 different executables. Because my engine is written completely in Rust, and installation/compilation is very easy with this language, it may be better to just write a one page compilation guide that will work everywhere.

[Ras]

I still wonder if these engines keep working if the kernel and/or glibc are updated on the Raspberry Pi.

Generally: yes. It's the other direction that often doesn't work, i.e. if you compile under a newer Linux, you cannot reliably expect the binary to run on an older Linux. That's why you compile on the oldest Linux that you intend to target.

Then again, having an easy way to build from source also allows the compiler to optimise for the CPU of the system if using -march=native. A big difference to Windows is that building from source is usually difficult under Windows, at least compared to Linux.