Game 7/10


In this game the French, Rubinstein variation was played until move 11.

All was level until around the half-way point. Here is the position after the CGP2024 moved 29…Ne7:

The CGP playing with the white pieces made a good move 30.Bxe6! If Black responds by taking white’s Bishop 30…KxB 31.Nxc5+ Ke5 and Black loses his Bishop. However, Black played 30…Ng6 and is a pawn down. Better would have been 30…Bxf3 31.gxf3 Kxe6 with a level position. The CGP2024 mis-calculated the position and not through a lack of search depth!

In the following position the CGP2024 made another bad move:

Black to play and a pawn down. Here the CGP2024 had not benefited from any extra endgame knowledge and made the move 33…g5? This weakens his Queen-side pawn structure after white plays 33.hxg6. White now has a winning chance.

However, as the game ended in a draw then obviously the original CGP must have also mis-calculated. In the following position white is in a good position with three passed pawns. Black does have a pawn on h3 but White can deal with that:

In this position the CGP turned a winning game into a draw with the move 47.Nd3? The game continued 47…h2 48.Nf2 to prevent Black queening 48…axb4 axb4 and now White cannot afford to move his knight and Black can hamper White’s pawn advance. A better move for White would have been 47.Ne4 maintaining the winning position. Play might have continued 47…h2 and now 48.Ng3+ is the best plan because Black cannot queen without losing it and he must move his King losing tempo:

Now White can move 49.b5 and Black is in trouble with passed pawns marching up the board!

I think in this case the CGP was unable to calculate far enough ahead to see the mistake and missed a winning chance.

Play battled on and I persevered in case of more mistakes but when the CGP2024 offered a draw I accepted on behalf of the CGP.

As expected they are level so far, although I really thought the newer model was going to lose this game.

3 ½ – 3 ½

Ray

Game 8/10

It’s another draw, this time by 3rd fold repition. I hope this is not boring for people, but I’m actually enjoying the test of the CG Pro engines at this slow time control. I knew the match would be close but wanted to try and find out if the differences in the engines would be highlighted. So far the CGPro2024 has a won game in the least number of moves, but this doesn’t mean much. The same could happen the other way around given enough games, and there is time yet to witness this.

The older CGPro does calculate further ahead, especially noticeable in the end game and I have seen it calculate an extra 2ply. That said, it’s not made a difference so far in the score in this match so maybe the extra knowledge in the newer CGPro makes up for this? Swings and roundabouts as they say, in some positions the extra search depth may help and in others, the extra knowledge may make the difference.



The Four Knights, symmetrical variation was played until both machines were out of book at move 10.

It’s been mentioned before but the Classic book (London) contains dubious lines and that is the case in this game. At move 9 both machines were still in book but the selected moves starting from Black’s 8th move are either interesting or dubious:

Black had just moved 8…Bd7?! Qe7 is considered good.

Play continued 9.Nd2 Ne7? Better would have been the book move 9…h6! 10.Bxf6 gxf6 11/Bxd7 Qxd7:

And white is in a much stronger position while Black’s King is now exposed.

Apart from weakness in the Classic opening book both machines played accurately and defended well, so we have the unsurprising draw.

Score now is 4 – 4

Ray

Game 9/10



Surprise, surprise another draw. They played the Pirc Defence: Austrian Attack and by move 19 were both out of book. There were no obvious blunders (only SF could see a slight advantage for White in places but neither machine would have seen that), in fact both played very accurately. So accurate that the game ended as a draw by 3-fold repetition.

Current score 4 ½ - 4 ½ With everything riding on the last game!

To be honest, if either machine wins the last game it’s not conclusive evidence that one is stronger than the other, probably more down to the luck of the chosen opening.

Ray

Die 2 Programme kennen sich, die spielen immer auf Remis

Yes, I think that must be the case

Ray

Game 10/10


The final game and unfortunately no win to make it more exciting finish. That said, at least in this match it’s showing that any strength difference between the machines is likely very small if at all, and may only be revealed after many more games. At least it’s not a backward step in terms of playing strength which I did not expect to witness anyway.

So on to the game:

The Caro-Kann: Exchange variation and Panov-Botvinnik Attack (wow!) was played until both were out of book at move 9.

Not much to be said, there were no blunders just good play from both sides.

Final score 5 – 5

They each achieved 2 won games and possibly the most exciting win was by the CGP2024 in game 4. Both the older CG Pro’s wins were playing with the black pieces.

I noticed that the older model’s eval would tend to be a little higher at times indicating a definite difference in the programs. I also felt that some of the attacking play by the programs was not analysed deep enough to be carried through with conviction. For example, in some situations one side would look to be forming an aggressive attack but would not have enough supporting pieces to see it through. This is very unlike the King engine but is a mistake I sometimes make in my own games so I should learn from this!

Ray

Interessante Matchs, der weiße Bauerndurchbruch des CGP 2024 in Game 4 war erstaunlich.

Interesting matches, CGP 2024's white pawn breakthrough in Game 4 was amazing.