Season 8 group 1a final results

Stage 1a is over, and the final results are:

The qualifiers are Komodo, Gull, Protector, Fire, Ginkgo and Texel. This is the same list as the one I predicted after 8 rounds, and indeed Jonny, Gaviota and Arasan could not catch up to the leaders. Jonny tried, winning all its last games, but the hole it was in was too deep to climb out of. The qualification was determined after 10 rounds, the last round was just for fun.

Arasan's last three games were draws against Ginkgo, Protector and Fire. All three are qualifiers - respect. I hope its ratings increases significantly, it also had a respectable performance in season 7 and it deserves a raise.

Komodo had two draws in the last three games, one of them against Exchess. Its only win was in the last round against Texel, and I feel that if Texel was not moving on the 30s increment for the last 35 moves it could have held a draw as well.

Ginkgo lost on time against Protector, in a balanced position after 19 moves.That's a total of 2 games with an unnatural ending in the stage, which is not too bad. Fortunately, these games did not affect qualification.

Notable game
Fruit-Gull, round 11
Game on TCEC archive

Fruit exchanged a rook for a bishop and pawn early in the game, but the game was judged to be balanced by both engines. Gull thought it had a small advantage after the queens and knights were removed:

Both sides cleared a few pawns, and by move 39 there was a rook exchange:

Now the strategy for Gull was clearer. The central white pawns could not be stopped but so was black's a4 pawn. Soon both sides queened and had their queen taken:

Now Gull would gladly exchange a rook for the bishop and central pawn, its A pawn will force the white king away from defending the king side, letting the black king to finish the game. I especially liked the position after move 52:

The pawn cannot advance, the bishop must stay to defend, and the rook cannot take the bishop. Black's king though has time to go to d6, allowing the pawn to be taken. Fruit decided not to exchange its bishop as well, but the game was over in any case. The amazing thing is that Gull completed this endgame without the use of tablebases.

Season 8 group 1a after 8 rounds

With 8 rounds one game away from being finished the stage 1a scoreboard is now

In rounds 6-8 there have been more white wins than black, and the statistics now look more reasonable, 30% draws, 40% white wins and 30% black wins.

Komodo, Gull and Ginkgo will surely qualify, and so will Fire if it stays stable (apparently its hash and thread number settings have been lowered). For the remaining two spots my prediction is Texel and Protector. Arasan's performace has been admirable, including a draw against Gull. this suggests that its rating should be higher, but it is not enough for qualification. Jonny needs to win all its remaining games, and I doubt it can with its performance so far. Gaviota has some chances, but taking into account that it has an extra point due to Fire's crash I think it won't make it.


First points for Exchess! it beat Fruit in the battle for last place, but it also beat Gaviota which was a bit surprising. Gaviota chose an inferior defense 1. e4 Nc6, and paid for it. Game was a bit wild with advanced passed pawns on both sides, Exchess was able to convert the win in the end.

Fire played a very long game against Komodo. On move 3 Fire thought for 20 minutes, again raising questions about stability and bugs. It did not crash though. The game had several stretches of shuffling but Komodo found a way to win. Could Fire have held if it had a little more time or if it was a bit more accurate?

Jonny started its game against Protector with 8 pawn moves, resulting in the following position:

Weird opening... The pawn wall quickly disintegrated and Protector was able to punish Jonny for this adventure.

Notable game
Jonny - Fire, round 8
Game on TCEC archive

This was the position after Jonny's 13th move:

Black's king could not castle and the g6 square was inviting a fork. Indeed, after a few moves white completed a forking combination taking the h8 rook for two knights.

Jonny's evaluation jumped over 2, Fire was much more conservative at 0.56, but which was more accurate? White managed to invade the 7th row, after queens and minor pieces were exchanged:

Fire and Jonny's evaluation did not change a lot. Despite the apparent danger to black's king, Fire managed to hold the position and force a rook exchange. In the ensuing endgame the pawns were gradually removed on both sides until the black bishop walked into a trap:

Fire simply traded the bishop for a pawn, and the game reached a drawn R vs N ending. So Fire was right all along.

Season 8 group 1a after 5 rounds

With 5 rounds over the group 1a scoreboard looks like this:

The ratings are starting to show, all the engines with elo>3000 are in the top half. Protector's situation improved considerably, including a draw against Komodo in a very long Q vs RN game. Jonny is only 1/5 but its losses are against top engines in this group and it may have a better record against easier opponents. Arasan is playing above its rating, though it is hard to see it qualify. Arasan lost to Komodo but it played well for 40 moves before Komodo could find a way to win.

The game that made the most noise was the Fire-Gaviota game in round 3. After a long battle Fire was clearly winning, even the win counter was close to adjudication, and then with 23 minutes on the clock Fire timed out and failed to make a move. This gave Gaviota an unexpected point, which explains its current place. Gaviota's ability to qualify is still in question. Fire didn't crash again yet, but it makes the viewers anxious every time it has a long think, including 16 minutes on move 1 in round 4. There's probably a bug somewhere, let's hope that Fire qualifies and gets a bugfix. It is a strong engine and it would be a shame to lose it on crashes.

In rounds 3-5 there have only been 3 (!) draws. The current draw rate is 30%, white wins 33% and black wins 37%. There have been a variety of openings according to ECO codes, for e4 games variations of the French are popular, and for d4 games variations of QGD. Only one Sicilian and one Ruy Lopez so far.

Notable game:
Round 3, Cheng - Komodo
Game on TCEC archive

On move 14 Komodo placed a strong knight on f4 making many threats on the king side.

Komodo expected Cheng to take the knight with its bishop, opening the g column for a rook. However, Cheng had other ideas, and it continued development on the queen side. Cheng continued to disregard Komodo's advice, ignoring the danger on its king side. On move 20 Komodo's evaluation started to jump. The knight was still at f4 but now the advice Komodo was giving did not invlove taking it since there were more urgent things to do.

Cheng's king side crumbled after several more moves and Komodo's pressure was converted to a win by move 34. The knight on f4 was still there (well, it moved and came back). Komodo: "I told you to take it, didn't I..."

Memorable Stockfish games, part 3

More memorable Stockfish games:
part 1
part 2

Season 6, stage 4, round 27, Stockfish - Houdini, Ra2
Game on TCEC archive

This game was played after half of stage 4 was over. After 8-RR Stockfish was at +6, Komodo was +4 and Houdini was +3. It was generally accepted that Stockfish was better, but Houdini was still a very strong opponent able to hold its ground, even against top engines.

After the 8 moves of the pre-selected opening each engine made its first move, and according to their evaluations there was nothing exciting going on. Then came Stockfish's 10th move:

Ra2. The crowd went slightly wild. The rook doesn't usually move forward so early in the game, and here Stockfish was hinting that it may use the empty second row to move a second rook to the king side quickly. The game continued with Stockfish preparing an attack on the king side and Houdini countering on the queen side. Houdini's evaluation remained drawish, it only started to sense the danger on move 18:

The rook is still on a2, waiting for its chance. In a few moves Stockfish opened the king side with a pawn attack, the evaluation skyrocketed, and then on move 25 the rook moved.

With so much firepower directed against its king, Houdini admitted defeat. The game was terminated on move 29 !!! Another WOW. In a match between two top engines this was almost unheard of. And of course there was Ra2. Ra2 was suggested in the chat as a move to change games around quickly, with laughs all around.

Season 8 group 1a after two rounds

Two rounds are already over, at a rate of about one round a day. Here is the scoreboard so far:

It's too early to think about qualifiers, Arasan and Gaviota with a better that expected result, while Protector and Jonny are diappointing so far. The openings have been diverse, and the draw rate so far is 50% with black winning 33% and white winning 17%. Letting the engines play from the first move is working well.

Jonny made a joke on move 1 of its game against Komodo. It started 1. e4, and evaluated this at 0.49! Komodo didn't laugh too much, it gradually turned the tables on Jonny, and converted a Q vs RB ending.  In Jonny's second game, black against Texel, after 1. d4 d5 Jonny evaluated 0.57, not so funny this time. I think the evaluation function needs to be examined.

In the Komodo-Gull game two of the favorites of this group met. The game ended in a draw, but there was one interesting event starting at move 35. Suddenly without warning Komodo's evaluation jumped for two moves, as if it saw a win. And then it returned to drawish values. But then Gull's evaluation did the same. According to this effect we can estimate that Komodo sees about two moves farther than Gull.

Notable game
Of the games I followed in these rounds the one I enjoyed the most was the Gaviota - Protector game.
Game on TCEC archive

Gaviota the underdog maintained a small positive score, and when it saw its chance it launched a furious pawn attack against black's king, here after move 30:

Just look at black's miserable bishop. With black's king stuck in the corner and the queen as its last defender, white launched an attack on the f7 pawn:

A final knight maneuver cracked black's defense for the win. Congratulations to Gaviota and its author Miguel Ballicora, always a welcome participant in the chat room with a lot of insightful commentary.
 

Season 8 is here !!!

After a long wait it's finally here !
It's really been too long, though now that it's here will I have time for other things?

There are some novelties this season:

• The tournament left Martin Thoresen's appartment and is now hosted by Chessdom. I noticed that they even changed the meaning of TCEC = TOP Chess Engine Championship. For me Thoresen will always be part of TCEC...
• New hardware with 20 cores. I'm not a hardware expert, the talk in the chat has been that although there are more cores than before they are slower, and the overall speed benefit is small. There are rumors that the final will be played with 40 cores, we'll see.
• Time control for the first stage is 90 minutes per engine plus 30 seconds per move. This will increase in the next stages by 30 minutes per stage, up to a monstrous 180 minutes per engine in the final. I think this is too long - every game in the final will take 7-8 hours.
• The final will consist of 100 games. At a rate of 3 games per day it will take a month to finish - I told you it was too long...
• The first stage will have no preset openings. I really enjoyed the no opening stage in last season, very happy they continued with this idea. The next stage will have short 2-move openings, and after that 8-move openings will return.

The first stage is divided into two groups. Group 1a contains Komodo, Gull, and Fire, group 1b will feature Stockfish and Houdini.

May the best engine win, and fun for all !!

Memorable Stockfish games, part 2

More memorable Stockfish games:
part 1
part 3

Season 6, stage 1b, round 6, Stockfish - Fire, the F stack
Game on TCEC archive


Stockfish and Fire were the highest rated engines in the first stage group 1b of season 6, which made this game one of the highlights of the stage.

From very early on in the game Stockfish had an evaluation higher than 1.5 and growing as the game progressed, while Fire's evaluation was around 0.7 and generally constant. Fire couldn't see its danger, and the question was how Stockfish will get the win. By move 45 Stockfish is a bishop up with almost no pawns but Fire still has a queen and rook, the winning path is not clear. Even though Stockfish's evaluation is over 4 it's principal variation does not look immediately winning. Checking a 6-man tablebase shows that if we remove the rooks and pawns the position is a draw. Is this what Fire is telling us?

After 11 more moves the game reaches this beautiful position which made this game so memorable to me:

Almost all pieces are stacked on the F column!! Stockfish sees a clear win now, but Fire still thinks its not in real danger. The illusion is broken on the next move, Fire's evaluation jumps to 4.77, and in a few more moves Fire loses its rook for a bishop and the game is over.

The analysis of this position is far beyond me. To my mind only engines can venture into such positions, and as we see in this game only the top ones manage to analyze them correctly.