New engine releases & news 2021

[j.t.]

Nalwald 1.9 Release.
Biggest change is that evaluation parameters are now optimized using generic gradient descent. Additionally, piece square tables are now dependent on the enemy and own king position (one piece square table for each of the 64 squares the enemy king could be on and one piece square table for each of the 64 squares our own king could be on). I also added (and removed) some other evaluation features.

Self-play improvement for 40 seconds + 0.3 seconds on a fast CPU is around 140 Elo, few tests against two other engines (Raven 1.10, Loki 3.0) indicate that true improvement is closer to 90 Elo.

[mvanthoor]

Release of Rustic Alpha 3.0.0

This release adds two new features: Killer Moves and Principal Variation Search. In self-play against the previous version Alpha 2, this new version is about +80 Elo (+/- 20 Elo) stronger. Since Alpha 3.0.0, a pre-compiled binary for MacOS (Intel) is also available.

New features:
- Killer Moves
- Principal Variation Search (PVS)
Changes:
- Switch versioning scheme to SemVer. Versions are going to be in the
form "a.b.c" from now on, with the following meaning:
- Increment **a**: A new strength-gaining feature was added.
- Increment **b**: A bug was fixed that gained strength.
- Increment **c**: A feature was added or a bug was fixed that did not
gain stregnth. It is not necessary to test such a version for a rating
change.
Misc:
- Updated crossbeam-channel to version 0.5.1
- A Makefile was added, so Rustic can be built using "GNU Make". When
typing "make" (or "gmake" in MacOS), the Makefile will build all Rustic
versions for the platform it's being compiled on.
- Re-add showing the size of the TT and number of threads in About.
- Fairly large update of the book on https://rustic-chess.org/.

This release has killer moves and PVS. Elo increase in self-play is +81, +/- 20 Elo. I'm satisfied if this version achieves +50 or +60 in the CCLR Blitz list over Alpha 2. Thanks in advance for the tests for the CCRL list

I also implemented and tested Aspiration Windows and History Heuristic, but they didn't gain any strength so I took them out again. I'll try again after the evaluation has improved in version 4.0.0.

The engine now has a Makefile. Just typing "make" (or "gmake" on MacOS, after you install GNU make through Homebrew) should be enough to build all Rustic CPU versions for your platform, assuming your development environment is installed correctly. The rest of the updates are the documentation for building Rustic (build.md), and the book/tutorial at https://rustic-chess.org.

The next version is going to be more exciting from a chess playing point of view. That will be the one with a tapered and tuned evaluation. I hope to hit at least 2100 with that version, and at if so, it will be Rustic 4.0.0, finally out of the "alpha" stage.

[Gabor Szots]

- Switch versioning scheme to SemVer. Versions are going to be in the
form "a.b.c" from now on, with the following meaning:
- Increment **a**: A new strength-gaining feature was added.
- Increment **b**: A bug was fixed that gained strength.
- Increment **c**: A feature was added or a bug was fixed that did not
gain stregnth. It is not necessary to test such a version for a rating
change.

And what does increment d mean?

[mvanthoor]

- Switch versioning scheme to SemVer. Versions are going to be in the
form "a.b.c" from now on, with the following meaning:
- Increment **a**: A new strength-gaining feature was added.
- Increment **b**: A bug was fixed that gained strength.
- Increment **c**: A feature was added or a bug was fixed that did not
gain stregnth. It is not necessary to test such a version for a rating
change.

And what does increment d mean?

It means that I implemented a feature which lost Elo, but got left in because of its mind-blowing cleverness

But a.b.c or x.y.z is a normal version scheme... MinimalChess (and many other engines) use it as well.

[supersharp77]

LC0 v0.28.0 rc-1 is out!! First release since Fefruary of 2021...

https://github.com/LeelaChessZero/lc0/r ... 0.28.0-rc1

Multigather is now made the default (and also improved). Some search settings
have changed meaning, so if you have modified values please discard them.
Specifically, max-collision-events, max-collision-visits and
max-out-of-order-evals-factor have changed default values, but other options
also affect the search. Similarly, check that your gui is not caching the old
values.
Performance improvements for the cuda/cudnn backends.
Support for policy focus during training.
Larger/stronger 15b default net for all packages except android, blas and dnnl
that get a new 10b network.
The distributed binaries come with the mimalloc memory allocator for better
performance when a large tree has to be destroyed (e.g. after an unexpected
move).
The legacy time manager will use more time for the first move after a long
book line.
The --preload command line flag will initialize the backend and load the
network during startup.
A 'fen' command was added as a UCI extension to print the current position.
Experimental onednn backend for recent intel cpus and gpus."

[jonkr]

Slow Chess 2.6 was released today.

Download from the website.

Also see the TalkChess Slow Chess thread for details or discussion.

[Rebel]

LC0 v0.28.0 rc-1 is out!! First release since Fefruary of 2021...

https://github.com/LeelaChessZero/lc0/r ... 0.28.0-rc1

Multigather is now made the default (and also improved). Some search settings
have changed meaning, so if you have modified values please discard them.
Specifically, max-collision-events, max-collision-visits and
max-out-of-order-evals-factor have changed default values, but other options
also affect the search. Similarly, check that your gui is not caching the old
values.
Performance improvements for the cuda/cudnn backends.
Support for policy focus during training.
Larger/stronger 15b default net for all packages except android, blas and dnnl
that get a new 10b network.
The distributed binaries come with the mimalloc memory allocator for better
performance when a large tree has to be destroyed (e.g. after an unexpected
move).
The legacy time manager will use more time for the first move after a long
book line.
The --preload command line flag will initialize the backend and load the
network during startup.
A 'fen' command was added as a UCI extension to print the current position.
Experimental onednn backend for recent intel cpus and gpus."

600 games

http://rebel13.nl/a/grl.htm

Will run for about a week.

Boring....

So far :

Code: Select all

Gambit Rating List
Running      : Gauntlet Lc0 v28
Time Control : Time control 120+2
Games        : 600

Results from file gauntlet-lc0-v28.pgn:

No. Name             Win Draw Loss Unf.  Score Games       %
------------------------------------------------------------
  1 Lc0 v28           +2   =3   -1   *0    3.5     6   58.3%
  2 sf13              +1   =0   -0   *0    1.0     1  100.0%
  3 Igel 3.0.5        +0   =1   -0   *0    0.5     1   50.0%
  4 Komodo-Dragon 2   +0   =1   -0   *0    0.5     1   50.0%
  5 Pedone 3.1        +0   =1   -0   *0    0.5     1   50.0%
  6 RubiChess 2.1     +0   =0   -1   *0    0.0     1    0.0%
  7 SlowChess 2.6     +0   =0   -1   *0    0.0     1    0.0%

Total Games:       6
White Wins:        2 (33.3%)
Black Wins:        1 (16.7%)
Draws:             3 (50.0%)
Unfinished:        0 (0.0%)

Estimated elo gain for Lc0_v28 
Elo pool : 3430
Lc0 v27 : 3529.0
Lc0_v28  : 3430.0
Difference : -99.0

[mar]

Chiron 5 seems to be out now
https://www.chironchess.com/news/

[CMCanavessi]

Chiron 5 seems to be out now
https://www.chironchess.com/news/

Looks like I will have to keep using Chiron 4 for my tournaments... 20 euros in my 3rd world country is a lot.

[Modern Times]

Well I started an FRC match earlier and went back to check it, and every single game terminated abnormally. Could be that I made a rare mistake, I'll re-check and try again.