New engine releases & news H1 2022

[jhonnold]

Berserk 9 is now available

Downloads available on GitHub as always: https://github.com/jhonnold/berserk/releases/tag/9

There are some interim results up on the CCRL FRC website -> +51 Elo so far.

http://ccrl.chessdom.com/ccrl/404FRC/ra ... _pure.html

Thank you for the fast testing!

Though I would like to apologize as the binaries I uploaded were far from optimal. I have uploaded new binaries which are significantly faster. A local test showed the `avx2` binaries are now ~20% faster which is extremely significant.

Code: Select all

       berserk-9-x64-avx2.exe       |     berserk-9-x64-avx2_old.exe     |
        mu              sigma       |        mu              sigma       |   Sp(1)/Sp(2)      3*sigma
------------------------------------+------------------------------------+------------------------------------
       2672985.000             0.000|       2233389.000             0.000|      19.683 %  +/-  0.000 %
       2672985.000             0.000|       2233899.500           721.956|      19.656 %  +/-  0.116 %
       2688347.000         26607.765|       2239211.000          9213.941|      20.056 %  +/-  2.082 %
       2696028.000         26607.765|       2241866.750          9209.226|      20.256 %  +/-  2.081 %
       2700636.600         25242.342|       2246797.800         13608.220|      20.198 %  +/-  1.844 %

[Modern Times]

Though I would like to apologize as the binaries I uploaded were far from optimal. I have uploaded new binaries which are significantly faster. A local test showed the `avx2` binaries are now ~20% faster which is extremely significant.

Code: Select all

       berserk-9-x64-avx2.exe       |     berserk-9-x64-avx2_old.exe     |
        mu              sigma       |        mu              sigma       |   Sp(1)/Sp(2)      3*sigma
------------------------------------+------------------------------------+------------------------------------
       2672985.000             0.000|       2233389.000             0.000|      19.683 %  +/-  0.000 %
       2672985.000             0.000|       2233899.500           721.956|      19.656 %  +/-  0.116 %
       2688347.000         26607.765|       2239211.000          9213.941|      20.056 %  +/-  2.082 %
       2696028.000         26607.765|       2241866.750          9209.226|      20.256 %  +/-  2.081 %
       2700636.600         25242.342|       2246797.800         13608.220|      20.198 %  +/-  1.844 %

I use berserk-9-x64-avx2-pext.exe

[jhonnold]

Though I would like to apologize as the binaries I uploaded were far from optimal. I have uploaded new binaries which are significantly faster. A local test showed the `avx2` binaries are now ~20% faster which is extremely significant.

Code: Select all

       berserk-9-x64-avx2.exe       |     berserk-9-x64-avx2_old.exe     |
        mu              sigma       |        mu              sigma       |   Sp(1)/Sp(2)      3*sigma
------------------------------------+------------------------------------+------------------------------------
       2672985.000             0.000|       2233389.000             0.000|      19.683 %  +/-  0.000 %
       2672985.000             0.000|       2233899.500           721.956|      19.656 %  +/-  0.116 %
       2688347.000         26607.765|       2239211.000          9213.941|      20.056 %  +/-  2.082 %
       2696028.000         26607.765|       2241866.750          9209.226|      20.256 %  +/-  2.081 %
       2700636.600         25242.342|       2246797.800         13608.220|      20.198 %  +/-  1.844 %

I use berserk-9-x64-avx2-pext.exe

The speed difference will be the same. This is just one example of how much worse the original binaries were unfortunately

[Modern Times]

Though I would like to apologize as the binaries I uploaded were far from optimal. I have uploaded new binaries which are significantly faster. A local test showed the `avx2` binaries are now ~20% faster which is extremely significant.

Code: Select all

       berserk-9-x64-avx2.exe       |     berserk-9-x64-avx2_old.exe     |
        mu              sigma       |        mu              sigma       |   Sp(1)/Sp(2)      3*sigma
------------------------------------+------------------------------------+------------------------------------
       2672985.000             0.000|       2233389.000             0.000|      19.683 %  +/-  0.000 %
       2672985.000             0.000|       2233899.500           721.956|      19.656 %  +/-  0.116 %
       2688347.000         26607.765|       2239211.000          9213.941|      20.056 %  +/-  2.082 %
       2696028.000         26607.765|       2241866.750          9209.226|      20.256 %  +/-  2.081 %
       2700636.600         25242.342|       2246797.800         13608.220|      20.198 %  +/-  1.844 %

I use berserk-9-x64-avx2-pext.exe

The speed difference will be the same. This is just one example of how much worse the original binaries were unfortunately

On my 5900X, the new exes are 6-7% faster, and on my i9 10900 they are 9-10% faster. Nowhere near the 20% difference you are seeing.
(go depth 24 as instructed on github)

[jhonnold]

Though I would like to apologize as the binaries I uploaded were far from optimal. I have uploaded new binaries which are significantly faster. A local test showed the `avx2` binaries are now ~20% faster which is extremely significant.

Code: Select all

       berserk-9-x64-avx2.exe       |     berserk-9-x64-avx2_old.exe     |
        mu              sigma       |        mu              sigma       |   Sp(1)/Sp(2)      3*sigma
------------------------------------+------------------------------------+------------------------------------
       2672985.000             0.000|       2233389.000             0.000|      19.683 %  +/-  0.000 %
       2672985.000             0.000|       2233899.500           721.956|      19.656 %  +/-  0.116 %
       2688347.000         26607.765|       2239211.000          9213.941|      20.056 %  +/-  2.082 %
       2696028.000         26607.765|       2241866.750          9209.226|      20.256 %  +/-  2.081 %
       2700636.600         25242.342|       2246797.800         13608.220|      20.198 %  +/-  1.844 %

I use berserk-9-x64-avx2-pext.exe

The speed difference will be the same. This is just one example of how much worse the original binaries were unfortunately

On my 5900X, the new exes are 6-7% faster, and on my i9 10900 they are 9-10% faster. Nowhere near the 20% difference you are seeing.
(go depth 24 as instructed on github)

Okay, I still think 6-7% is pretty significant, but that makes me less worried about my original binaries which looked atrocious. Thanks for checking the speeds of the old vs new.

[Guenther]

JackyChess 0.12.0
https://github.com/matthiaslang/jackychess

[akanalytics]

Odonata 0.6 released...

Changelog
* Pawn structure examined during eval
* Eval caching
* Promos accounted for in q-search
* Internal search metrics

About 150 Elo gain in self play, that never seems to translate into anything remotely similar in real-life play, sadly!

Andy

https://github.com/akanalytics/odonata/ ... s/tag/v0.6

[Damir]

Gargantua

Version 1.0
Thanks to the Author Claudio M. Camacho mkd dance4

UCI chess engine with NNUE evaluation .

Download : https://github.com/mkd/gargantua/releases/tag/1.0

Homepage / Instructions : https://github.com/mkd/gargantua

Fully Bitboard-based implementation
Pure NNUE evaluation
Negamax with quiescent search
Basic UCI protocol support
Cross-compilation for Win64 platforms
Principal Search Variation
Late Move Reductions (LMR)
Null moving pruning
Aspiration window
Basic time controls
Transposition tables
Contempt factor
Razoring
Mate distance pruning
Futility pruning
Reverse futility prunning
Late move pruning
No-hashmove reduction

[Damir]

This seems to be a single core engine which makes it useless... Only 1 param called Contempt and nothing else...

[jmcd]

Clovis v1.0
My second attempt at a chess engine!
https://github.com/jonathanmcdermid/Clo ... es/tag/1.0