Last Update: 27.01.2023
Dear Chess Friends,
As usual, I am pleased to announce that,Here is another new NNUE competition!And organized in the name of Waldemar!
1st of all,All NNUE engines are tested again with new opening suite!E.g created in same system way as previous 'Unique' vers. What's new, based on more openings...total 149 openingsFor anyone missed, White + Black managed to win beforeBy engines 3700+..yes... this is a new concept of testing!And via this method, I believe that it's more fair.. right?Btw, the current Unique vers. produces minus plus 10%Less draws than Balsa suite..sure not so strong as Balsa,But if nothing else...at least more fun for sure...!)
Note also that,I used a new Banksia 0.58 GUI and frankly as far as I noticed, It's really great one..includes a lot of futures, quite suitable for Concurrent matches too, BIG thanks to Banksia's programmer
What I can say more, I tested DarkSister again, reason simply to see it is stable or not..e.g On my older 2x E5-2686 + Cutechess GUI: DarkiSister was crashing.. But under these new hardware conditions.. no any crash is recorded... Btw I used some older versions which were Winners in previous tours: FisherovMonk and Zeus, but the rest participated ones are quite new!
By the way, StockfishMZ test is re-started... e.g in previous hardware setup: StockfishMZ was crashing, but with other Bios settings, now is ok... And for solution I set: ACPI SRAT L3 Cache as NUMA Domain = Auto E.g before (when StockfishMZ's crashes) this option was Enabled
Meanwhile, once more (for anyone missed, interested...) Why I prefer 2 Cores/Game + Blitz, the main reason, otherwise: Just in case let's say 32 Cores + Blitz, Rapid etc. then almost all Games are expecting to be ended as draw, very likely over 99%+. In other words, I may loose my interest..e.g if almost all results will Be equal...then what is the point ? on other hand, Sure, with very weak openings, we can reduce overall draws..but For nowadays I have no such plans, thanks for your understanding..
And here are the final results, yes under new machine (2x EPYC 7B12) Kookaburra managed to be Number One !! SF-PB via latest update: 2nd Not so bad performance comes by DarkSister,Genko, Polyfish, Zeus etc...
The Winner Kookaburra - My Congratulations to K. and Stockfish team!
Rank Name Elo + - Games Score Oppo. Draws 1 Kookaburra v1.03a 3770 8 9 560 52% 3757 87% 2 SF PB 040123 3768 9 9 560 52% 3757 88% 3 Dark Sister 1.9a 3766 9 9 560 51% 3757 90% 4 Genko 1.1 3765 9 9 560 51% 3757 89% 5 Polyfish 230115 3764 9 9 560 51% 3757 91% 6 Zeus 26.5 3763 9 9 560 51% 3757 87% 7 Eman 8.60 3761 9 9 560 51% 3757 90% 8 Stockfish 140123 3759 9 9 560 50% 3757 86% 9 CorChess 140123 3759 9 9 560 50% 3757 88% 10 Brainlearn 21.1 3758 9 9 560 50% 3757 88% 11 Charisma Diamond 3754 9 9 560 50% 3758 88% 12 FisherovMonk 1.2 3752 9 9 560 49% 3758 89% 13 Blue Marlin 15.6 3752 9 9 560 49% 3758 85% 14 ProteusSF-P 15.1 3750 9 9 560 49% 3758 86% 15 StockfishMZ 051222 3720 8 7 560 43% 3760 81%
• Calculated by Bayeselo / ProteusSF is fixed to 3750 Elo / Blue Name = Last Update
More Details: • Almost all of the NNUE engines are AVX2, tested via theirs default Evalfiles Exception, Zeus v25.6 is played via nn.zeus, which is based on13.07.2022 • Tech (average nodes, depths, time/m per move, others per game): Results • The overall draw percentage (based on 4200 games) is not so high: 87% • All the played tour games are available (in PGN): NNUE XXIV / StockfishMZ
Best, Sedat
Dear Chess Friends,
As usual, I am pleased to announce that,Here is another new NNUE competition!And organized in the name of Waldemar!
1st of all,All NNUE engines are tested again with new opening suite!E.g created in same system way as previous 'Unique' vers. What's new, based on more openings...total 149 openingsFor anyone missed, White + Black managed to win beforeBy engines 3700+..yes... this is a new concept of testing!And via this method, I believe that it's more fair.. right?Btw, the current Unique vers. produces minus plus 10%Less draws than Balsa suite..sure not so strong as Balsa,But if nothing else...at least more fun for sure...!)
Note also that,I used a new Banksia 0.58 GUI and frankly as far as I noticed, It's really great one..includes a lot of futures, quite suitable for Concurrent matches too, BIG thanks to Banksia's programmer
What I can say more, I tested DarkSister again, reason simply to see it is stable or not..e.g On my older 2x E5-2686 + Cutechess GUI: DarkiSister was crashing.. But under these new hardware conditions.. no any crash is recorded... Btw I used some older versions which were Winners in previous tours: FisherovMonk and Zeus, but the rest participated ones are quite new!
By the way, StockfishMZ test is re-started... e.g in previous hardware setup: StockfishMZ was crashing, but with other Bios settings, now is ok... And for solution I set: ACPI SRAT L3 Cache as NUMA Domain = Auto E.g before (when StockfishMZ's crashes) this option was Enabled
Meanwhile, once more (for anyone missed, interested...) Why I prefer 2 Cores/Game + Blitz, the main reason, otherwise: Just in case let's say 32 Cores + Blitz, Rapid etc. then almost all Games are expecting to be ended as draw, very likely over 99%+. In other words, I may loose my interest..e.g if almost all results will Be equal...then what is the point ? on other hand, Sure, with very weak openings, we can reduce overall draws..but For nowadays I have no such plans, thanks for your understanding..
And here are the final results, yes under new machine (2x EPYC 7B12) Kookaburra managed to be Number One !! SF-PB via latest update: 2nd Not so bad performance comes by DarkSister,Genko, Polyfish, Zeus etc...
The Winner Kookaburra - My Congratulations to K. and Stockfish team!
Rank Name Elo + - Games Score Oppo. Draws 1 Kookaburra v1.03a 3770 8 9 560 52% 3757 87% 2 SF PB 040123 3768 9 9 560 52% 3757 88% 3 Dark Sister 1.9a 3766 9 9 560 51% 3757 90% 4 Genko 1.1 3765 9 9 560 51% 3757 89% 5 Polyfish 230115 3764 9 9 560 51% 3757 91% 6 Zeus 26.5 3763 9 9 560 51% 3757 87% 7 Eman 8.60 3761 9 9 560 51% 3757 90% 8 Stockfish 140123 3759 9 9 560 50% 3757 86% 9 CorChess 140123 3759 9 9 560 50% 3757 88% 10 Brainlearn 21.1 3758 9 9 560 50% 3757 88% 11 Charisma Diamond 3754 9 9 560 50% 3758 88% 12 FisherovMonk 1.2 3752 9 9 560 49% 3758 89% 13 Blue Marlin 15.6 3752 9 9 560 49% 3758 85% 14 ProteusSF-P 15.1 3750 9 9 560 49% 3758 86% 15 StockfishMZ 051222 3720 8 7 560 43% 3760 81%
• Calculated by Bayeselo / ProteusSF is fixed to 3750 Elo / Blue Name = Last Update
More Details: • Almost all of the NNUE engines are AVX2, tested via theirs default Evalfiles Exception, Zeus v25.6 is played via nn.zeus, which is based on13.07.2022 • Tech (average nodes, depths, time/m per move, others per game): Results • The overall draw percentage (based on 4200 games) is not so high: 87% • All the played tour games are available (in PGN): NNUE XXIV / StockfishMZ
Best, Sedat