Chess324

[Modern Times]

it seems to be 29% http://ccrl.chessdom.com/ccrl/404FRC/

Yes it is, but that is single CPU blitz, plus there are a lot of games played where the Elo difference between the engines is very large, up to +/- 300 Elo and slightly more in some cases. That was necessary sometimes simply because there would be too small a number of opponents otherwise, if you restricted it to less then that. So not comparable to TCEC or CCC at all.

[Chessqueen]

it seems to be 29% http://ccrl.chessdom.com/ccrl/404FRC/

Yes it is, but that is single CPU blitz, plus there are a lot of games played where the Elo difference between the engines is very large, up to +/- 300 Elo and slightly more in some cases. That was necessary sometimes simply because there would be too small a number of opponents otherwise, if you restricted it to less then that. So not comparable to TCEC or CCC at all.

Plus the Draws average of the top 8 engines on this CCRL FRC list is = 48.00 using single CPU Blitz

[Chessqueen]

it seems to be 29% http://ccrl.chessdom.com/ccrl/404FRC/

Yes it is, but that is single CPU blitz, plus there are a lot of games played where the Elo difference between the engines is very large, up to +/- 300 Elo and slightly more in some cases. That was necessary sometimes simply because there would be too small a number of opponents otherwise, if you restricted it to less then that. So not comparable to TCEC or CCC at all.

My conclusion is that the more threads and more powerful computer the more draws you will get, no matter what variant engine vs engine are playing, In another 5 years even Chess324 with more Powerchips or powerful computer most of the variants will have at least 95% Draws

[lkaufman]

So far ...
Draws : 194/275 (70.5 %)

With time at top level engines may hit draw death 90%, but i suspect no more.

What about the percentage of draws at Chess960 from CCRL games ?

Sorry, not familiar with CCRL games.

it seems to be 29% http://ccrl.chessdom.com/ccrl/404FRC/

There is no comparison between CCRL games run on 1 or 4 threads vs CCC or TCEC games run on hundreds of threads. There was a CCC chess960 tournament about a year ago that would be a fair comparison; I don't remember the draw percentage in the RR, but in the SF vs Dragon final it was nearly 99% draws out of 600 games! Maybe someone can figure out how to look up the RR event. However, I ran my own matches under identical conditions to determine this. Latest Dragon dev. vs sf 15, 2' + 1", one thread. Playing chess324, the draw percentage was 72.2% for the complete 648 games. Playing chess960, it is about 3/4 complete and so far the draw percentage is 80.9%. Also, the elo gap is over 50% larger in chess324 due to the fewer draws. So chess324 clearly reduces the draws compared to chess960. Probably the effect will be larger with more threads and more time, since with FRC it apparently approaches 100% draws while with chess324 a fair number of positions are near the win/draw line so there should always be some wins at any time control or thread count.

[Lazy_Frank]

So far ...
Draws : 194/275 (70.5 %)

With time at top level engines may hit draw death 90%, but i suspect no more.

What about the percentage of draws at Chess960 from CCRL games ?

Sorry, not familiar with CCRL games.

it seems to be 29% http://ccrl.chessdom.com/ccrl/404FRC/

There is no comparison between CCRL games run on 1 or 4 threads vs CCC or TCEC games run on hundreds of threads. There was a CCC chess960 tournament about a year ago that would be a fair comparison; I don't remember the draw percentage in the RR, but in the SF vs Dragon final it was nearly 99% draws out of 600 games! Maybe someone can figure out how to look up the RR event. However, I ran my own matches under identical conditions to determine this. Latest Dragon dev. vs sf 15, 2' + 1", one thread. Playing chess324, the draw percentage was 72.2% for the complete 648 games. Playing chess960, it is about 3/4 complete and so far the draw percentage is 80.9%. Also, the elo gap is over 50% larger in chess324 due to the fewer draws. So chess324 clearly reduces the draws compared to chess960. Probably the effect will be larger with more threads and more time, since with FRC it apparently approaches 100% draws while with chess324 a fair number of positions are near the win/draw line so there should always be some wins at any time control or thread count.

https://www.chess.com/computer-chess-ch ... hip-finals
https://www.chess.com/computer-chess-ch ... semifinals
https://www.chess.com/computer-chess-ch ... ampionship

https://cccfiles.chess.com/archive/tour ... 236923.pgn
https://cccfiles.chess.com/archive/tour ... 235721.pgn
https://cccfiles.chess.com/archive/tour ... 234510.pgn

Invite to CCC Discord channel:
https://discord.gg/nzchhnnZ

[Chessqueen]

So far ...
Draws : 194/275 (70.5 %)

With time at top level engines may hit draw death 90%, but i suspect no more.

What about the percentage of draws at Chess960 from CCRL games ?

Sorry, not familiar with CCRL games.

it seems to be 29% http://ccrl.chessdom.com/ccrl/404FRC/

There is no comparison between CCRL games run on 1 or 4 threads vs CCC or TCEC games run on hundreds of threads. There was a CCC chess960 tournament about a year ago that would be a fair comparison; I don't remember the draw percentage in the RR, but in the SF vs Dragon final it was nearly 99% draws out of 600 games! Maybe someone can figure out how to look up the RR event. However, I ran my own matches under identical conditions to determine this. Latest Dragon dev. vs sf 15, 2' + 1", one thread. Playing chess324, the draw percentage was 72.2% for the complete 648 games. Playing chess960, it is about 3/4 complete and so far the draw percentage is 80.9%. Also, the elo gap is over 50% larger in chess324 due to the fewer draws. So chess324 clearly reduces the draws compared to chess960. Probably the effect will be larger with more threads and more time, since with FRC it apparently approaches 100% draws while with chess324 a fair number of positions are near the win/draw line so there should always be some wins at any time control or thread count.

That is the reason why I came to this conclusion My conclusion is that the more threads and more powerful computer the more draws you will get, no matter what variant engine vs engine are playing, In another 5 years even Chess324 with more Powerchips or powerful computer most of the variants will have close to 95% of draws .

[lkaufman]

So far ...
Draws : 194/275 (70.5 %)

With time at top level engines may hit draw death 90%, but i suspect no more.

What about the percentage of draws at Chess960 from CCRL games ?

Sorry, not familiar with CCRL games.

it seems to be 29% http://ccrl.chessdom.com/ccrl/404FRC/

There is no comparison between CCRL games run on 1 or 4 threads vs CCC or TCEC games run on hundreds of threads. There was a CCC chess960 tournament about a year ago that would be a fair comparison; I don't remember the draw percentage in the RR, but in the SF vs Dragon final it was nearly 99% draws out of 600 games! Maybe someone can figure out how to look up the RR event. However, I ran my own matches under identical conditions to determine this. Latest Dragon dev. vs sf 15, 2' + 1", one thread. Playing chess324, the draw percentage was 72.2% for the complete 648 games. Playing chess960, it is about 3/4 complete and so far the draw percentage is 80.9%. Also, the elo gap is over 50% larger in chess324 due to the fewer draws. So chess324 clearly reduces the draws compared to chess960. Probably the effect will be larger with more threads and more time, since with FRC it apparently approaches 100% draws while with chess324 a fair number of positions are near the win/draw line so there should always be some wins at any time control or thread count.

That is the reason why I came to this conclusion My conclusion is that the more threads and more powerful computer the more draws you will get, no matter what variant engine vs engine are playing, In another 5 years even Chess324 with more Powerchips or powerful computer most of the variants will have close to 95% of draws .

You could be right about chess324, but your statement is not correct as written. If the initial positions in some variant are actually winning for one side, for example starting the game with White missing c2 pawn, then as computers get stronger the draw percentage will shrink, not grow. Similarly if we introduce tie-breaking rules (for example, in case of a draw counting material and pawn advancement to break the tie, or mobility) then White's advantage may become winning and again the draw percentage will shrink with more firepower. Ideally, if a variant has variety and leads to positions so close to the win/draw line that only a perfect player could tell whether they are won or lost, then the draw percentage should hover near 50% until computers approached perfection. Not easy to make a variant like that with simple rules, easy enough if you can pick positions to play based on engine scores.

[lkaufman]

I noticed that in the ongoing chess.com Chess324 event, looking only at the games between leaders Stockfish and Dragon, four out of fifteen have been decisive so far, 27%. For comparison, in the previous final match between the same two engines (earlier versions) in chess960, only 11 out of 600 were decisive, 2%! What an enormous difference! Note also that neither Stockfish nor Dragon has lost a single game to any of the other six engines in the current event out of 176 such games, suggesting that even if a few of the 324 positions are "winning" for White, they are not easily winning.

[Chessqueen]

I noticed that in the ongoing chess.com Chess324 event, looking only at the games between leaders Stockfish and Dragon, four out of fifteen have been decisive so far, 27%. For comparison, in the previous final match between the same two engines (earlier versions) in chess960, only 11 out of 600 were decisive, 2%! What an enormous difference! Note also that neither Stockfish nor Dragon has lost a single game to any of the other six engines in the current event out of 176 such games, suggesting that even if a few of the 324 positions are "winning" for White, they are not easily winning.

Ethereal is showing a great performance in Chess324, it is holding 3rd place after Dragon 3.1, I suppose it has a small net that does not waste too much time looking for unknown positions that it has not been trained for like LCO and great evaluation https://www.chess.com/computer-chess-championship

Note: Can we play Chess324 on Chess.com ?

[AndrewGrant]

I noticed that in the ongoing chess.com Chess324 event, looking only at the games between leaders Stockfish and Dragon, four out of fifteen have been decisive so far, 27%. For comparison, in the previous final match between the same two engines (earlier versions) in chess960, only 11 out of 600 were decisive, 2%! What an enormous difference! Note also that neither Stockfish nor Dragon has lost a single game to any of the other six engines in the current event out of 176 such games, suggesting that even if a few of the 324 positions are "winning" for White, they are not easily winning.

Ethereal is showing a great performance in Chess324, it is holding 3rd place after Dragon 3.1, I suppose it has a small net that does not waste too much time looking for unknown positions that it has not been trained for like LCO and great evaluation https://www.chess.com/computer-chess-championship

Note: Can we play Chess324 on Chess.com ?

Ethereal is particularly good at FRC and FRC-like play. I estimate Ethereal as only -70 to Dragon in FRC, although much more in Standard play.