SlowChess Blitz Classic 2.0

[Rebel]

Running the gauntlet for the GRL.

Games : 2000

Elo Pool : 3462

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

The battle for the 4th place.

Actually the 3th place as Lc0 v27 ran on superior hardware.

[peter]

You can download Slow Chess 2.7 from the webpage :
https://www.3dkingdoms.com/chess/slow.htm

• Standard chess is improved by about +35 Elo compared to 2.6.(10000 bullet game gauntlet against similar strength engines, 8-move book.)
• FRC play is improved by +115 Elo (in a 3840-game bullet match against 2.6 )
• I restarted a new opening book so it should have fewer bad lines, but still not a competitive book.
• As usual, the only net compatible with 2.7 is the net that comes with it. (Eventually this may change but I'm always fiddling with my net structures and it isn't based on anything else.)

If you want to read in more detail about changes this version, check this thread : http://talkchess.com/forum3/viewtopic.p ... 92#p902881

Thanks for the new version!
To me SlowChess has the best mate- search besides the specialised programs like Chest and Gustav. As for long- line- mates it has replaced SF Matefinder in my usage at forward- backward- analysis of such puzzles and positions regards

[CMCanavessi]

You can download Slow Chess 2.7 from the webpage :
https://www.3dkingdoms.com/chess/slow.htm

• Standard chess is improved by about +35 Elo compared to 2.6.(10000 bullet game gauntlet against similar strength engines, 8-move book.)
• FRC play is improved by +115 Elo (in a 3840-game bullet match against 2.6 )
• I restarted a new opening book so it should have fewer bad lines, but still not a competitive book.
• As usual, the only net compatible with 2.7 is the net that comes with it. (Eventually this may change but I'm always fiddling with my net structures and it isn't based on anything else.)

If you want to read in more detail about changes this version, check this thread : http://talkchess.com/forum3/viewtopic.p ... 92#p902881

I noticed that "SyzygyProbeDepth" default value was changed from 3 in 2.6 to 2 in 2.7. Why was this changed and how does this affect gameplay while using 6-men Syzygy?

[Thomas Lagershausen]

Great

What a fast development

Thx a lot.

[Rebel]

Slowchess 2.7 +16 elo

Details at - https://prodeo.actieforum.com/t358p25-g ... sults#4265

Maybe more luck at other lists.

[jonkr]

Thanks for the new version!
To me SlowChess has the best mate- search besides the specialised programs like Chest and Gustav. As for long- line- mates it has replaced SF Matefinder in my usage at forward- backward- analysis of such puzzles and positions regards

I'm glad some people are noticing Slow is good at mate-finding. There are a few special things it does like priority for keeping mates in transposition table and near leaves of searches it doesn't prune mate threats or dangerous checks. I used to test the mate-finding more than I do now. I might start testing stuff like mate-finding again as I think I'm reaching limits in strength increase. (I can probably catch up to Stockfish 11 in strength, but current SF is in another league entirely than I can reach.)

Great

What a fast development

Thx a lot.

Thanks.
I would have probably waited longer before releasing but I've wanted to fix the FRC play for a long while and this version should.

I noticed that "SyzygyProbeDepth" default value was changed from 3 in 2.6 to 2 in 2.7. Why was this changed and how does this affect gameplay while using 6-men Syzygy?

It probably makes no noticeable difference. I tested no depth limit on Syzygy probing in main search and everything ran fine, so I was thinking of removing the setting entirely, but I have an SSD so maybe on a slow hard-disk it might be needed.

Slowchess 2.7 +16 elo

Details at - https://prodeo.actieforum.com/t358p25-g ... sults#4265

Maybe more luck at other lists.

While the gambit list performance is a bit disappointing, with different openings, opponents and random variation it's on the lower end of expectations but not surprising. I should try testing gambit openings too next time.

What I do wonder about is the strong performance of Stockfish in your lists. Does your test machine support AVX2? If not I suppose that's a large part of it. (In my personal tests Slow 2.7 narrowly beat Stockfish 10 in 3 out of 4 standard opening books I tried, in FRC it's still quite a bit weaker though even with the improvement in 2.7)

[jonkr]

What I do wonder about is the strong performance of Stockfish in your lists. Does your test machine support AVX2? If not I suppose that's a large part of it.

Just to clarify this I mean Stockfish 9 and 10, the new Stockfish I do expect to be far above other programs as they are.

[AlexChess]

You can download Slow Chess 2.7 from the webpage :
https://www.3dkingdoms.com/chess/slow.htm

• Standard chess is improved by about +35 Elo compared to 2.6.(10000 bullet game gauntlet against similar strength engines, 8-move book.)
• FRC play is improved by +115 Elo (in a 3840-game bullet match against 2.6 )
• I restarted a new opening book so it should have fewer bad lines, but still not a competitive book.
• As usual, the only net compatible with 2.7 is the net that comes with it. (Eventually this may change but I'm always fiddling with my net structures and it isn't based on anything else.)

If you want to read in more detail about changes this version, check this thread : http://talkchess.com/forum3/viewtopic.p ... 92#p902881

Thank you!

[Rebel]

What I do wonder about is the strong performance of Stockfish in your lists. Does your test machine support AVX2? If not I suppose that's a large part of it.

Just to clarify this I mean Stockfish 9 and 10, the new Stockfish I do expect to be far above other programs as they are.

Code: Select all

No. Name                 Win Draw Loss Unf.  Score Games       %
----------------------------------------------------------------
  1 SlowChess 2.7       +401 =1069 -530   *0  935.5  2000   46.8%
  2 Stockfish 14        +124  =76   -0   *0  162.0   200   81.0%
  3 Komodo-Dragon 2      +99  =97   -4   *0  147.5   200   73.8%
  4 Stockfish 10         +53 =119  -28   *0  112.5   200   56.2%
  5 Ethereal 13.25-NNUE  +49 =118  -33   *0  108.0   200   54.0%
  6 Stockfish 9          +46 =112  -42   *0  102.0   200   51.0%
  7 Komodo 14            +45 =101  -54   *0   95.5   200   47.8%
  8 h6                   +40 =110  -50   *0   95.0   200   47.5%
  9 RubiChess 2.2        +33 =118  -49   *0   92.0   200   46.0%
 10 Pedone 3.1           +21 =110  -69   *0   76.0   200   38.0%
 11 Igel 3.0.5           +20 =108  -72   *0   74.0   200   37.0%

1. The PC's I use for the GRL support AVX but not AVX2, is the difference really that big?

2. As you can see from the table Slowchess played only 200 games against each engine.

3. Note the GRL is a tad different than other rating lists due to the nature of the openings, see the CCRL-GRL comparison I recently did.

4. As you already suggested just test with the 100 gambit positions so now and then.

[jonkr]

1. The PC's I use for the GRL support AVX but not AVX2, is the difference really that big?

2. As you can see from the table Slowchess played only 200 games against each engine.

3. Note the GRL is a tad different than other rating lists due to the nature of the openings, see the CCRL-GRL comparison I recently did.

4. As you already suggested just test with the 100 gambit positions so now and then.

Testing on my computer (Ryzen 9 3950x), the NPS is about 33% faster with AVX2 instructions than with just the SSE2 ones. I never measured how much Elo that is, but it is a large enough difference I was curious about AVX2 support. (Also I never tried to measure details for different computer types, instructions sets, and expected speeds. I just made the AVX2 path since that instruction set is a big jump in speed. )

Openings can also can make a large difference, especially with neural nets, FRC being the most extreme example. So probably a combination of the 2.