Latest Update: 15.04.2022
Dear Chess Friends,
Btw, here are available some new testings...actually I ran many.. But I realized to publish those standings, which are useful ones... For example, in latest days I've played over than 15.000 games... And in these testings, some engines are more drawish, weaker etc. In short, after spending several testing days, Now is more clear regarding to determine which cond. is best..)
Ok... and here are the mentioned new testings:
NNUE vs NON-NNUE Test (in Auto-232 Mode):
1 NNUE ON 1 Core 1m+1s +22/-20/=166 50.48% 105.0/208 2 NNUE OFF 6 Core 3m+1s +20/-22/=166 49.52% 103.0/208
Some Notes: The current test is played as Ponder OFF under Arena 2.0.1 GUIs NNUE ON = Komodo Dragon 2.6 / NNUE OFF = Cfish 050521 For comparing I used same Cfish ver (e.g played in many book tours) As you may see, Komodo Dragon played with 1 Core and 1min+1sec Where Cfish NON-NNUE is played with full 6 Cores and 3 min+1 sec Komodo is played with 64 MB hash, Cfish played with 512 MB hashThe overall draw percentage (based on 208 games) not so high: 80% Both chess engines are used to play via Perfect 2022 opening book it seems just under these cond. both players are very close in strength In other words, the planning next tournament games will be not so bad!---------------------------------------------------------------------------------------------
HashTable Test: 64 MB vs 128 MB
1 64 mb hash +13/-9/=178 51.00% 102.0/2002 128 mb hash +9/-13/=178 49.00% 98.0/200
Note: The above Hashtable test is played by Komodo Dragon 2.6 BIG sizes don't mean better..it all depends on our used conditions---------------------------------------------------------------------------------------------
Cfish vs Komodo Test: Elo difference is just 1 Elo
1 Cfish 200621 +64/-62/=674 50.13% 401.0/8002 Komodo Dragon 2.6 +62/-64/=674 49.88% 399.0/800
More Details:It seems both NNUE engines are almost identical in strength:.Great!The overall draw percentage (based on 800 games) not so high: 84%Frankly, I can not remember to be so strong with such low draw ratio..E.g other 3700+ NNUE engines produce minus plus 90-95% drawsSure, I mean in case of using strong short openings for both sidesBut via slower time controls, more cores, ponder on as enabled etc...Then, you may know: to appear a 'win' will be just as dream...really !)For this reason, I preferred to use as 1 Core and Bullet Time controlBoth testings played under Cutechess Qt 5.11.1 with TC: 1m + 0.5sAs opening book, Balsa suite is used (tested with reversed colors)Download all the played games (in PGN format): NNUE Testings Best, Sedat
Dear Chess Friends,
As usual, I am pleased to announce,
Here is another new NNUE strength testing, organized for: Results
For anyone missed, interested:
Some time ago, I ran similar book tourney, for previous: Standings Where this time I will test the opening books by Komodo Dragon 2.6 Vs Cfish 200621 NNUE, which is very close to Komodo's strength And in this way, there will be no matches with same kind of Evalfies In other words, my target is simply to reduce the overall draw ratio... And in same time my goal is to produce better and quality games... Btw, if NNUE as MP or 1 core does not make BIG sense: Results As I mentioned before and to be more clear, I'm not against NNUE testings, however the real problem is that Mostly Top engines+ Evalfiles produce very high draw numbers Sure it depends on cond. but usually almost all games are draws Plus many critical, weak opening lines can be ended as draws too Normally I like draws..but when too much: Impressive, but Boring! Anyhow the good news is that, e.g under current test conditions: Overall draw ratio is approx.10% less...what we can need more? Otherwise, the life wouldn't be boring if we do the same things? Note that according to my latest new NNUE testings, E.g via Balsa openings: overall draw ratio is minus plus 80-85% Frankly, these numbers are not high...because I used as NNUE.. Not sure what will be the draws via deep books, but we will see.. As other test, 64 MB hash performed slightly better than 128 MB.. Later, I hope to test some deep books via NON-NNUE engine too In this way, we will have possibility to see the influence as well.. Note also that, I don't plan to use the latest SF NNUE engines.. Due to, the overall draws are expecting to be over than 95%... You may know, all these tours are just for fun, no more, no less... So once more, please don't take too seriously the next results... Thanks for your understanding...Btw, here are available some new testings...actually I ran many.. But I realized to publish those standings, which are useful ones... For example, in latest days I've played over than 15.000 games... And in these testings, some engines are more drawish, weaker etc. In short, after spending several testing days, Now is more clear regarding to determine which cond. is best..)
Ok... and here are the mentioned new testings:
NNUE vs NON-NNUE Test (in Auto-232 Mode):
1 NNUE ON 1 Core 1m+1s +22/-20/=166 50.48% 105.0/208 2 NNUE OFF 6 Core 3m+1s +20/-22/=166 49.52% 103.0/208
Some Notes: The current test is played as Ponder OFF under Arena 2.0.1 GUIs NNUE ON = Komodo Dragon 2.6 / NNUE OFF = Cfish 050521 For comparing I used same Cfish ver (e.g played in many book tours) As you may see, Komodo Dragon played with 1 Core and 1min+1sec Where Cfish NON-NNUE is played with full 6 Cores and 3 min+1 sec Komodo is played with 64 MB hash, Cfish played with 512 MB hashThe overall draw percentage (based on 208 games) not so high: 80% Both chess engines are used to play via Perfect 2022 opening book it seems just under these cond. both players are very close in strength In other words, the planning next tournament games will be not so bad!---------------------------------------------------------------------------------------------
HashTable Test: 64 MB vs 128 MB
1 64 mb hash +13/-9/=178 51.00% 102.0/2002 128 mb hash +9/-13/=178 49.00% 98.0/200
Note: The above Hashtable test is played by Komodo Dragon 2.6 BIG sizes don't mean better..it all depends on our used conditions---------------------------------------------------------------------------------------------
Cfish vs Komodo Test: Elo difference is just 1 Elo
1 Cfish 200621 +64/-62/=674 50.13% 401.0/8002 Komodo Dragon 2.6 +62/-64/=674 49.88% 399.0/800
More Details:It seems both NNUE engines are almost identical in strength:.Great!The overall draw percentage (based on 800 games) not so high: 84%Frankly, I can not remember to be so strong with such low draw ratio..E.g other 3700+ NNUE engines produce minus plus 90-95% drawsSure, I mean in case of using strong short openings for both sidesBut via slower time controls, more cores, ponder on as enabled etc...Then, you may know: to appear a 'win' will be just as dream...really !)For this reason, I preferred to use as 1 Core and Bullet Time controlBoth testings played under Cutechess Qt 5.11.1 with TC: 1m + 0.5sAs opening book, Balsa suite is used (tested with reversed colors)Download all the played games (in PGN format): NNUE Testings Best, Sedat