New engine releases 2019

[JohnWoe]

Thanks for the builds!
This new version(1.11 / 1.12) is just an eval cleanup. There was stuff I wrote long time ago I didn't understand so I removed them.
It seems Sapeli plays better.
+ It castles ASAP.
+ It keeps pawn in front of his king.
I need to run more tests before making a release.

[flok]

Embla 2.0.6

- null move improvement
- tt improvement

https://vanheusden.com/Embla/files/Embl ... -win64.zip

[Stan Arts]

Nemeton 1.81 from the link in my signature below.

Kind regards,
Stan

[Gabor Szots]

Nemeton 1.81 from the link in my signature below.

Kind regards,
Stan

Thank you, Stan. Judging by the version number the changes are minor. What are they?

[Stan Arts]

Nemeton 1.81 from the link in my signature below.

Kind regards,
Stan

Thank you, Stan. Judging by the version number the changes are minor. What are they?

Hi Gabor,

Usual code optimisations but the main thing is a nullmove trick. I reduce an extra ply if static eval is significantly out of window.

Cheers,
Stan

[Gabor Szots]

Nemeton 1.81 from the link in my signature below.

Kind regards,
Stan

Thank you, Stan. Judging by the version number the changes are minor. What are they?

Hi Gabor,

Usual code optimisations but the main thing is a nullmove trick. I reduce an extra ply if static eval is significantly out of window.

Cheers,
Stan

Thank you. I'm going to start its testing on Sunday, if everything goes as planned.

Cheers,

[voffka]

Igel 1.6.0 at https://github.com/vshcherbyna/igel/releases/tag/1.6.0 (binaries for Linux and Windows)

What's new:

- Re-write killer moves functionality and drop mate killers
- Re-write move ordering and introduce counter move heuristics
- Re-write LMR functionality: take into account 'improving' factor and prune more aggressively
- Increase MAX_PLY to 128 from 64
- Use tt moves in qsearch as a first move

Strength improvements over 1.5.1 in CCRL 40/40 time control:

Code: Select all

Score of Igel 1.6.0 64 POPCNT vs Igel 1.5.1 64 POPCNT: 20 - 4 - 60  [0.595] 84
Elo difference: 67.00 +/- 38.62

Relative strength compared to division 6 engines in CCRL 40/4 time control:

Code: Select all

Rank Name                          Elo     +/-   Games   Score   Draws
   0 Igel 1.6.0 64 POPCNT           91      22     770   62.8%   23.0%
   1 Francheska                     14      58     100   52.0%   28.0%
   2 Donna 4.1                     -14      60     100   48.0%   24.0%
   3 ECE-X3.5 64Bit                -24      59     100   46.5%   27.0%
   4 Minic 0.60                    -74      60     100   39.5%   27.0%
   5 WyldChess                    -100      62     100   36.0%   24.0%
   6 DanaSah 7.3                  -164      67     100   28.0%   20.0%
   7 K2 v.0.91                    -182      68     100   26.0%   20.0%
   8 Coiled 0.6 x64               -301     111      70   15.0%   10.0%

[xr_a_y]

Igel 1.6.0 at https://github.com/vshcherbyna/igel/releases/tag/1.6.0 (binaries for Linux and Windows)

What's new:

- Re-write killer moves functionality and drop mate killers
- Re-write move ordering and introduce counter move heuristics
- Re-write LMR functionality: take into account 'improving' factor and prune more aggressively
- Increase MAX_PLY to 128 from 64
- Use tt moves in qsearch as a first move

Strength improvements over 1.5.1 in CCRL 40/40 time control:

Code: Select all

Score of Igel 1.6.0 64 POPCNT vs Igel 1.5.1 64 POPCNT: 20 - 4 - 60  [0.595] 84
Elo difference: 67.00 +/- 38.62

Relative strength compared to division 6 engines in CCRL 40/4 time control:

Code: Select all

Rank Name                          Elo     +/-   Games   Score   Draws
   0 Igel 1.6.0 64 POPCNT           91      22     770   62.8%   23.0%
   1 Francheska                     14      58     100   52.0%   28.0%
   2 Donna 4.1                     -14      60     100   48.0%   24.0%
   3 ECE-X3.5 64Bit                -24      59     100   46.5%   27.0%
   4 Minic 0.60                    -74      60     100   39.5%   27.0%
   5 WyldChess                    -100      62     100   36.0%   24.0%
   6 DanaSah 7.3                  -164      67     100   28.0%   20.0%
   7 K2 v.0.91                    -182      68     100   26.0%   20.0%
   8 Coiled 0.6 x64               -301     111      70   15.0%   10.0%

Wow, at faster TC this is not the same

Code: Select all

Rank Name                                Elo     +/-   Games   Score   Draws
   1 zurichess-neuchatel-linux-amd64     264      16    2177   82.1%   16.4%
   2 MinkoChess_1.3_x64                  239      16    2178   79.9%   16.7%
   3 ruy-1.1.9-linux                      93      13    2178   63.1%   21.1%
   4 redqueen-1.1.98.linux64              -2      13    2177   49.7%   22.0%
 >>5 minic_dev_linux_x64                 -10      13    2177   48.6%   24.5%
   6 wyldchess1.51_bmi                   -21      13    2177   47.0%   23.4%
   7 minic_0.53_linux_x64_avx2_bmi2      -32      13    2177   45.5%   23.9%
   8 minic_0.58_linux_x64_avx2_bmi2      -38      13    2177   44.5%   25.2%
   9 GreKo-Linux-64                      -49      13    2178   42.9%   18.1%
  10 Fridolin310                         -81      13    2178   38.6%   21.2%
>>11 igel-x64_popcnt1.5.1               -112      14    2178   34.5%   18.9%
  12 asymptote-v0.4.2                   -202      15    2178   23.8%   14.7%

13065 of 66000 games finished.  

Do you know what is causing this difference between long and short TC in Igel ?

[voffka]

Salut Vivien,

[...]

Do you know what is causing this difference between long and short TC in Igel ?

I presume two factors: time management and LMR. Also, in long tc there is a chance tablebase probing code will kick-in. I will check it out, but since most of my testing time controls are 40/4 and 40/40 this is not so urgent

Mimic does not support tablebases yet?

P.S. I see you are testing a Linux version of Igel. Actually, most of my testing is done on Windows, so perhaps there is something in Linux version that slows it down.

[xr_a_y]

Salut Vivien,

[...]

Do you know what is causing this difference between long and short TC in Igel ?

I presume two factors: time management and LMR. Also, in long tc there is a chance tablebase probing code will kick-in. I will check it out, but since most of my testing time controls are 40/4 and 40/40 this is not so urgent

Mimic does not support tablebases yet?

P.S. I see you are testing a Linux version of Igel. Actually, most of my testing is done on Windows, so perhaps there is something in Linux version that slows it down.

Indeed, I often test under Linux.

Minic has experimental support of syzygy table but not really tested yet ... I'll work on that ... just finished KPK...