Saturday, March 21, 2020

Season 17 premier division, CPU only, after RR2

The organizers have decided to order the premier division games so that the GPU hardware usage is optimized. The CPU only games will be played first without a need for a GPU machine. Then all the CPU-GPU and GPU-GPU games will be played. Eventually all games will be played (a 2xDRR tournament), but it will be very difficult to understand the relative standings of the engines until the very end.

For me this is a little disappointing, I think that the most interesting part of this division is the CPU-GPU clash and now it will be postponed and the crosstables will not be balanced for most of the division. It will be very surprising to me if Stockfish does not dominate the CPU part of the division. I would like to see whether Leela can put up a fight this season, whether AllieStein is as strong as last season, and whether Stoofvlees can beat AllieStein or Leela.

Despite these reservations, the chess is going to be of the highest level as it always is in TCEC !!

Crosstable after RR2


As expected Stockfish is in the lead with 3 wins, all when playing white. Fire is in second place with a win and no loss, quite surprising for Fire, though it may be temporary and there are many games to be played yet.
Houdini is last with 2 losses, Houdini is using only 64 threads of a possible 176 and this may affect its strength relative to the other engines.

Interesting games

Game 1, Stockfish - Ethereal: The game reached a RB vs RB position on move 33, both evals were close to 1 and Stockfish quickly went up a pawn and created a passer. Ethereal had to block the passer and Stockfish used this to capture a second pawn. Ethereal tried to counter by pushing a passer to the 2nd rank, Stockfish forced a rook exchange and the game was adjudicated.

Game 6, Stockfish - Komodo: Stockfish had a small eval advantage after the opening, which started to increase after 25 moves to over 1. By move 40 Stockfish was a pawn up with a pawn majority on the king side. After exchanging most pieces the game reached a QB vs QB position, Stockfish's eval climbed rapidly while Komodo's eval stayed around 2. The endgame was long and subtle, Komodo got the pawn back and tried to create a fortress. The white king assisted in blocking a black pawn and finally Komodo acknowledged that it was losing. The game was adjudicated with Stockfish 2 pawns up in a queen ending.

Game 9, Fire - KomodoMCTS: see featured game below.

Game 12, Komodo - Houdini: Komodo gave a pawn and opened a file on the king side, after exchanging queens it had a passer on the 6th rank facing the black king. A strong bishop on the long diagonal added pressure on the black king and evals increased over 2. Houdini had to give a rook for a knight, the engines exchanged down to a RR vs RB position. The black rook had to block the passer, Komodo started to pick up pawns and the game was adjudicated.

Game 23, Stockfish - Houdini: Houdini captured a pawn with its queen and opened a file for Stockfish facing the black king. Stockfish then gave a rook for a knight while threatening mate. Evals jumped as Stockfish increased the pressure, Houdini exchanged queens and saved its king, but Stockfish captured a rook and was a piece up. Houdini tried to defend for another 15 moves before giving up.

Featured game: Fire - KomodoMCTS
Premier division, game 9
Link to game on TCEC

KomodoMCTS opened the queen side early, its queen went forward and Fire used its pawns to push it back. Evals climbed to around 1 after Fire captured a rook for a bishop and pawn, and then placed its knight on c5.


KomodoMCTS exchanged knights on c5, at the cost of a white passer. Fire pushed the passer to the 7th rank while evals increased slowly.


KomodoMCTS tried to counter on the king side, it pushed its e passer forward and attacked with its queen. Fire kept its king safe, it exchanged pieces and traded its c passer for another on the a file.


KomodoMCTS had to abandon its passer to block the white a pawn, then lost its bishop to capture it. Fire was a rook up and the game was adjudicated a few moves later.

2 comments:

  1. The original RR1-3 finish after 30, 60 and 90 games in 'phase 2' - a good time to assess the ranking.

    ReplyDelete
    Replies
    1. I agree, every 30 games. I won't be doing any "if we look only at the original RR1 games the table would look like this" though.

      Delete