Texel 1.10

[petero2]

Version 1.10 of my chess engine Texel is now available for download.

This version is around 170 elo stronger than 1.09 at hyper-bullet self play. I believe this corresponds to around 120 elo on the CCRL 40/15 list.

The only changes affecting strength for this release are faster neural network evaluation and using a better network trained on data generated by Texel 1.09.

The same pre-compiled binaries as for 1.09 are included. There are no pre-built 32-bit binaries, but 32-bit is still supported in the source code and build system.

[Dariusz]

Peter, thx for the new Texel, good progress! (MCERL ongoing, MacBook Pro M1).

86 Texel 1.10 : 3388 1039.5 1532 67.9%
87 HIARCS 15.2 : 3388 976.0 1676 58.2%
88 Caissa 1.6.3 (Wine) : 3387 634.0 1044 60.7%
89 RubiChess 2021 : 3386 261.5 449 58.2%
90 PowerFritz 18 (Wine) : 3386 909.0 1830 49.7%
91 Uralochka v3.37c (Wine) : 3385 437.5 980 44.6%
92 Halogen 11 : 3385 710.5 1423 49.9%
93 Viridithas 8.0.0 : 3380 889.5 1880 47.3%
94 Weiss 2.1-dev : 3371 701.0 1324 52.9%
95 Velvet 5.1.0 : 3369 960.0 1941 49.5%
96 Komodo 13.02 : 3365 645.0 1366 47.2%
97 Arasan 24.0 : 3361 1082.0 1876 57.7%
98 Ethereal 13.50 : 3361 756.0 1497 50.5%
99 HIARCS 15.1 : 3360 1063.5 2197 48.4%
100 Tucano 11.00 : 3352 1095.0 1503 72.9%
101 Arasan 23.4.0-43-g5968e0b : 3351 1228.0 2196 55.9%
102 HIARCS 15.0 : 3350 896.5 1792 50.0%
103 Stormphrax 1.0.0 : 3342 737.0 1296 56.9%
104 Velvet 5.0.0 : 3340 681.5 1188 57.4%
105 Xiphos 0.6 : 3338 653.5 1269 51.5%
106 Frozenight 6.0.0 : 3336 829.5 1516 54.7%
107 Smallbrain 6.0 : 3323 510.0 1133 45.0%
108 Ethereal 13.00 : 3316 323.5 670 48.3%
109 Komodo 14.1 MCTS : 3314 734.0 1282 57.3%
110 Alexandria 4.0 : 3301 850.5 1547 55.0%
111 Stash 34.0 : 3300 544.5 1140 47.8%
112 Komodo 12.1.1 : 3298 465.5 806 57.8%
113 Viridithas 7.0.0 : 3285 706.0 1374 51.4%
114 Pawn 1.0 : 3285 1068.0 2247 47.5%
115 Velvet 4.1.0 : 3283 389.0 930 41.8%
116 Tenax 0.7.9 (Wine) : 3283 562.5 1138 49.4%
117 Alexandria 3.0-dev : 3282 306.0 676 45.3%
118 Lc0 v0.29.0 (791556) : 3280 1361.5 2841 47.9%
119 rofChade 2.3 : 3274 432.0 1004 43.0%
120 Laser 1.7 : 3272 535.5 1167 45.9%
121 Texel 1.09 : 3271 1049.5 2186 48.0%

[petero2]

Hi Darius,

Thanks for testing Texel on macOS. I don't have a mac myself to test on, but the following CMake settings are supposed to be available:

USE_CTZ
USE_NEON
USE_NEON_DOT
USE_PREFETCH

I think all of them should work on the MacBook Pro M1. Are those settings available for you and does it work to enable them? (If USE_NEON_DOT works there is no need to also enable USE_NEON, but it does not hurt either.)

[Dann Corbit]

Thanks for releasing it.
I have downloaded the binary and source and look forward to great pleasure in watching and learning.

[Dann Corbit]

Configuration question:
I have the full set of Gaviota 5 man files and the full set of Syzygy 6 man files, with most important 7 man files installed.
Currently, I have informed Texel of the location of both file sets.
Should I limit it to one of them?

[petero2]

Configuration question:
I have the full set of Gaviota 5 man files and the full set of Syzygy 6 man files, with most important 7 man files installed.
Currently, I have informed Texel of the location of both file sets.
Should I limit it to one of them?

Using both TBs at the same time should be best. Texel will intelligently decide when to probe from each table. There is for example no need to probe a DTM/DTZ table unless the position has already been proven to be won/lost. (Unless the corresponding WDL table is missing.)

Texel assumes that Syzygy WDL probe is fastest, followed by Syzygy DTZ probe, and that Gaviota DTM probe is slowest. This will usually be the case if the underlying storage is reasonably configured.

This post might be of interest: http://talkchess.com/forum3/viewtopic.php?p=679321

[Scally]

Hi Peter,

Compiled and running fine on Picochess on my 32 bit Raspberry Pi …

Thanks,

Al

[Dariusz]

Hi Darius,

Thanks for testing Texel on macOS. I don't have a mac myself to test on, but the following CMake settings are supposed to be available:

USE_CTZ
USE_NEON
USE_NEON_DOT
USE_PREFETCH

I think all of them should work on the MacBook Pro M1. Are those settings available for you and does it work to enable them? (If USE_NEON_DOT works there is no need to also enable USE_NEON, but it does not hurt either.)

It works, I can compile Texel 1.10.

cmake -USE_CTZ -USE_NEON -USE_NEON_DOT -USE_PREFETCH ..
-- The CXX compiler identification is AppleClang 15.0.0.15000040
-- The C compiler identification is AppleClang 15.0.0.15000040
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /Library/Developer/CommandLineTools/usr/bin/c++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /Library/Developer/CommandLineTools/usr/bin/cc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Setting build type: Release
-- Performing Test supported_-Wno-misleading-indentation
-- Performing Test supported_-Wno-misleading-indentation - Success
-- Performing Test supported_-Wno-unused-result
-- Performing Test supported_-Wno-unused-result - Success
-- Performing Test supported_-Wno-psabi
-- Performing Test supported_-Wno-psabi - Success
-- Performing Test supported_-fno-stack-protector
-- Performing Test supported_-fno-stack-protector - Success
-- librt not found.
-- Could NOT find Armadillo (missing: ARMADILLO_INCLUDE_DIR)
-- Could NOT find GSL (missing: GSL_INCLUDE_DIR GSL_LIBRARY) (found version "")
-- Found PkgConfig: /opt/homebrew/bin/pkg-config (found version "0.29.2")
-- Checking for module 'gtkmm-3.0'
-- No package 'gtkmm-3.0' found
Not building bookgui
-- Could NOT find Torch (missing: Torch_DIR)
Not building torchutil
CMake Deprecation Warning at test/gtest/CMakeLists.txt:56 (cmake_minimum_required):
Compatibility with CMake < 3.5 will be removed from a future version of
CMake.

Update the VERSION argument <min> value or use a ...<max> suffix to tell
CMake that the project does not need compatibility with older versions.


CMake Warning (dev) at test/gtest/cmake/internal_utils.cmake:243 (find_package):
Policy CMP0148 is not set: The FindPythonInterp and FindPythonLibs modules
are removed. Run "cmake --help-policy CMP0148" for policy details. Use
the cmake_policy command to set the policy and suppress this warning.

Call Stack (most recent call first):
test/gtest/CMakeLists.txt:91 (include)
This warning is for project developers. Use -Wno-dev to suppress it.

-- Found PythonInterp: /opt/homebrew/bin/python3.11 (found version "3.11.5")
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Success
-- Found Threads: TRUE
-- Configuring done (2.4s)
-- Generating done (0.1s)
-- Build files have been written to: /Users/chessengeria/Downloads/texel-1.10/build

[petero2]

I don't have a mac myself to test on, but the following CMake settings are supposed to be available:

USE_CTZ
USE_NEON
USE_NEON_DOT
USE_PREFETCH

I think all of them should work on the MacBook Pro M1. Are those settings available for you and does it work to enable them? (If USE_NEON_DOT works there is no need to also enable USE_NEON, but it does not hurt either.)

It works, I can compile Texel 1.10.

cmake -USE_CTZ -USE_NEON -USE_NEON_DOT -USE_PREFETCH ..
...

Actually cmake does not complain if you specify non-existing settings like that. You can run this instead:

Code: Select all

cmake -DUSE_CTZ=on -DUSE_PREFETCH=on -DUSE_NEON=on -DUSE_NEON_DOT=on ..

Alternatively you can first run "cmake .." without extra arguments and then run "cmake-gui ." and check in the GUI what settings are available.

To check if the settings are effective, I suggest running texel from the command line and search to depth 14 from the starting position:

Code: Select all

./texel
go depth 14

Then look at the last reported nps value and compare with/without using NEON/NEON_DOT.

[Dariusz]

I don't have a mac myself to test on, but the following CMake settings are supposed to be available:

USE_CTZ
USE_NEON
USE_NEON_DOT
USE_PREFETCH

I think all of them should work on the MacBook Pro M1. Are those settings available for you and does it work to enable them? (If USE_NEON_DOT works there is no need to also enable USE_NEON, but it does not hurt either.)

It works, I can compile Texel 1.10.

cmake -USE_CTZ -USE_NEON -USE_NEON_DOT -USE_PREFETCH ..
...

Actually cmake does not complain if you specify non-existing settings like that. You can run this instead:

Code: Select all

cmake -DUSE_CTZ=on -DUSE_PREFETCH=on -DUSE_NEON=on -DUSE_NEON_DOT=on ..

Alternatively you can first run "cmake .." without extra arguments and then run "cmake-gui ." and check in the GUI what settings are available.

To check if the settings are effective, I suggest running texel from the command line and search to depth 14 from the starting position:

Code: Select all

./texel
go depth 14

Then look at the last reported nps value and compare with/without using NEON/NEON_DOT.

with using NEON/NEON_DOT.
info depth 14
info depth 14 score cp 30 time 503 nodes 1055600 nps 2098608 pv e2e4 e7e6 d2d4 d7d5 e4e5 c7c5 c2c3 b8c6 g1f3 g8e7 f1d3 e7g6 e1g1 f8e7 f1e1 c5c4
info nodes 1247829 nps 2104264 hashfull 592 time 593
bestmove e2e4 ponder e7e6


without using NEON/NEON_DOT.
info depth 14
info depth 14 score cp 30 time 970 nodes 1055600 nps 1088247 pv e2e4 e7e6 d2d4 d7d5 e4e5 c7c5 c2c3 b8c6 g1f3 g8e7 f1d3 e7g6 e1g1 f8e7 f1e1 c5c4
info currmove c2c4 currmovenumber 3
info currmove e2e3 currmovenumber 4
info currmove h2h3 currmovenumber 5
info currmove g2g3 currmovenumber 6
info currmove a2a3 currmovenumber 7
info currmove b2b3 currmovenumber 8
info currmove b1c3 currmovenumber 9
info currmove d2d3 currmovenumber 10
info currmove g1f3 currmovenumber 11
info currmove a2a4 currmovenumber 12
info currmove h2h4 currmovenumber 13
info currmove b2b4 currmovenumber 14
info currmove f2f4 currmovenumber 15
info currmove b1a3 currmovenumber 16
info currmove c2c3 currmovenumber 17
info currmove g1h3 currmovenumber 18
info currmove g2g4 currmovenumber 19
info currmove f2f3 currmovenumber 20
info nodes 1247829 nps 1086959 hashfull 592 time 1148
bestmove e2e4 ponder e7e6


With NEON/NEONDOT it is about twice as fast!
Thanks for the compilation tips