Last Update: 21.02.2023
Dear Chess Friends,
As usually, I am pleased to announce that, Another new series of NNUE tournaments...
Some Notes, As you may see, this time I picked Top 7 players of recent tour Results What's new (comparing with the previous one): 5m+2s, 1024 MB hash
And what a pity that, Koivisto 9.2 seems to be buggy..e.g several games are lost due to as White/Black makes an illegal move..I hope next Koivisto to be stable ! And I have a small request from all programmers(before final releases): I suggest serious own beta testings..for example, in this tourney, some Players are gained very easy points...and this is not so fair..thank you...
What I can say more, The used time controls 2m+1s and 5m+2s did not make BIG influences Sure for a better conclusion more games are required, but it seems that Lc0 is doing slightly better via slower TC... actually this is known issue... And plus in this championship too, Lc0 played as concurrent games: 3
Meanwhile, if we are going to compare the current blitz conditions with myOlder tours (on 2x E5-2686), these games can be counted close to Rapid!
And the Winner is again Stockfish - My Congratulations to Stockfish team!
Rank Name Elo + - games score oppo. draws 1 Stockfish 15.1 3812 11 11 300 60% 3758 78% 2 Raubfisch X48e 3800 11 12 300 57% 3760 79% 3 KomodoDragon 3.2 3785 12 12 300 54% 3762 77% 4 Lc0 v030 809261 3770 12 12 300 51% 3765 81% 5 Fat Fritz 2 3755 12 12 300 48% 3767 76% 6 Berserk 11 3737 12 11 300 44% 3770 79% 7 Koivisto 9.2 3700 11 10 300 36% 3776 69%
• Calculated by Bayeselo / Raubfisch fixed to 3800 Elo / Blue Name = Last Update
More Details: • Almost all of NNUE engines are AVX2, tested via theirs default Evalfiles • As alternative (plus here) Arena 3.5.1 is used (e.g Lc0 engine is played) • The overall draw percentage (based on 1050 games) is not so high 77% Btw, the overall draws are increased (e.g about 20% than previous tour) • All the played tour games are available (in PGN format): NNUE CS XXVI
Best, Sedat
Dear Chess Friends,
As usually, I am pleased to announce that, Another new series of NNUE tournaments...
Some Notes, As you may see, this time I picked Top 7 players of recent tour Results What's new (comparing with the previous one): 5m+2s, 1024 MB hash
And what a pity that, Koivisto 9.2 seems to be buggy..e.g several games are lost due to as White/Black makes an illegal move..I hope next Koivisto to be stable ! And I have a small request from all programmers(before final releases): I suggest serious own beta testings..for example, in this tourney, some Players are gained very easy points...and this is not so fair..thank you...
What I can say more, The used time controls 2m+1s and 5m+2s did not make BIG influences Sure for a better conclusion more games are required, but it seems that Lc0 is doing slightly better via slower TC... actually this is known issue... And plus in this championship too, Lc0 played as concurrent games: 3
Meanwhile, if we are going to compare the current blitz conditions with myOlder tours (on 2x E5-2686), these games can be counted close to Rapid!
And the Winner is again Stockfish - My Congratulations to Stockfish team!
Rank Name Elo + - games score oppo. draws 1 Stockfish 15.1 3812 11 11 300 60% 3758 78% 2 Raubfisch X48e 3800 11 12 300 57% 3760 79% 3 KomodoDragon 3.2 3785 12 12 300 54% 3762 77% 4 Lc0 v030 809261 3770 12 12 300 51% 3765 81% 5 Fat Fritz 2 3755 12 12 300 48% 3767 76% 6 Berserk 11 3737 12 11 300 44% 3770 79% 7 Koivisto 9.2 3700 11 10 300 36% 3776 69%
• Calculated by Bayeselo / Raubfisch fixed to 3800 Elo / Blue Name = Last Update
More Details: • Almost all of NNUE engines are AVX2, tested via theirs default Evalfiles • As alternative (plus here) Arena 3.5.1 is used (e.g Lc0 engine is played) • The overall draw percentage (based on 1050 games) is not so high 77% Btw, the overall draws are increased (e.g about 20% than previous tour) • All the played tour games are available (in PGN format): NNUE CS XXVI
Best, Sedat