Latest Update: 05.04.2021
Hello Chess Friends
I am pleased to announce that,Recently I organized another new TC competitions...Yes...a new idea: Time Control testings, but via NNUE...Note also that from time to time, I run similar competitions!For example, anyone interested can check older: 2014 / 2019
And meanwhile, once more I'd like to mention that,We noticed that via NNUE, the Elo's progress is HUGE!Less blunder moves and better quality games for sure!About NNUE draw percentage...this time no comments )On other hand, if we are going to compare NON-NNUEVia deep opening books the Elo gain is BIG too, like NNUE!E.g in LI, Houdini 6.03 via deep books performed much better!
In short,Any engine performance is depending on our used conditions!Btw, in case of NON-NNUE testings:Deep vs Deep books: the overall draw percentage is higher too!E.g, in case of tests via short openings: approx. 10 - 15% less...
And now let's see the 'Time Control' influences, 1st of all,Definitely for a better conclusion more games are required!However, I think that the current new data is not so bad...And can be useful...if nothing else at least for my side !)So with the current new test results:- The time control's Elo difference depends...For example, under Blitz conditions: 2m+1s vs 8m+1sThe results are minus plus 23 Elo in favor for 8m+1sSo, it's clear that no much Elo gain via EvalFiles...
1st Test: 2min+1sec vs 8min+1sec 1 8min+1sec +13/-2/=153 53.27% 89.5/1682 2min+1sec +2/-13/=153 46.73% 78.5/168
Overall Draw Percentage (based on 168 games): 91%
But in case of Duel match: Bullet 20s+1s vs Blitz 2m+1sThen the results are approx. 70 Elo in favor Blitz 2m+1sSo it seems that, (under Bullet conditions):More blunder moves and NN strength suffers for sure!
2nd Test: 20sec+1sec vs 2min+1sec
1 2min+1sec +43/-0/=175 59.86% 130.5/2182 20sec1sec +0/-43/=175 40.14% 87.5/218
Overall Draw Percentage (based on 218 games): 80%
More Details,All of the games are played manly by CF EXT 291120 NNUEAs alternative, Stockfish 13 NNUE engine is used to play tooBut the matches are played: CF EXT vs CF EXT or SF vs SFBoth chess engines are played via same nn-62ef826d1a6d.nnueAuto232 matches: Arena 3.5.1 and ShredderClassic GUIs usedCurrent TC testings are played on same hardware: 2x E5-2686Perfect 2021 opening book played as Book Learning disabledDownload all the played games (in PGN) SCCT TC Testings II
After all, this is true that,Via NNUE, with slower time controls or giving more cores etc.Does not make BIG sense as before...!To be more clear, as references/sources:Exc. that the hardware speeds do not make so BIG role: ResultsPlus in this test: not so much Elo gain 3 Cores vs 1 Core: ResultsFor anyone missed, as NON-NNUE, Elo diff.is HUGE: Results
And as usual,Many thanks for your interest...
Greetings,Sedat
Hello Chess Friends
I am pleased to announce that,Recently I organized another new TC competitions...Yes...a new idea: Time Control testings, but via NNUE...Note also that from time to time, I run similar competitions!For example, anyone interested can check older: 2014 / 2019
And meanwhile, once more I'd like to mention that,We noticed that via NNUE, the Elo's progress is HUGE!Less blunder moves and better quality games for sure!About NNUE draw percentage...this time no comments )On other hand, if we are going to compare NON-NNUEVia deep opening books the Elo gain is BIG too, like NNUE!E.g in LI, Houdini 6.03 via deep books performed much better!
In short,Any engine performance is depending on our used conditions!Btw, in case of NON-NNUE testings:Deep vs Deep books: the overall draw percentage is higher too!E.g, in case of tests via short openings: approx. 10 - 15% less...
And now let's see the 'Time Control' influences, 1st of all,Definitely for a better conclusion more games are required!However, I think that the current new data is not so bad...And can be useful...if nothing else at least for my side !)So with the current new test results:- The time control's Elo difference depends...For example, under Blitz conditions: 2m+1s vs 8m+1sThe results are minus plus 23 Elo in favor for 8m+1sSo, it's clear that no much Elo gain via EvalFiles...
1st Test: 2min+1sec vs 8min+1sec 1 8min+1sec +13/-2/=153 53.27% 89.5/1682 2min+1sec +2/-13/=153 46.73% 78.5/168
Overall Draw Percentage (based on 168 games): 91%
But in case of Duel match: Bullet 20s+1s vs Blitz 2m+1sThen the results are approx. 70 Elo in favor Blitz 2m+1sSo it seems that, (under Bullet conditions):More blunder moves and NN strength suffers for sure!
2nd Test: 20sec+1sec vs 2min+1sec
1 2min+1sec +43/-0/=175 59.86% 130.5/2182 20sec1sec +0/-43/=175 40.14% 87.5/218
Overall Draw Percentage (based on 218 games): 80%
More Details,All of the games are played manly by CF EXT 291120 NNUEAs alternative, Stockfish 13 NNUE engine is used to play tooBut the matches are played: CF EXT vs CF EXT or SF vs SFBoth chess engines are played via same nn-62ef826d1a6d.nnueAuto232 matches: Arena 3.5.1 and ShredderClassic GUIs usedCurrent TC testings are played on same hardware: 2x E5-2686Perfect 2021 opening book played as Book Learning disabledDownload all the played games (in PGN) SCCT TC Testings II
After all, this is true that,Via NNUE, with slower time controls or giving more cores etc.Does not make BIG sense as before...!To be more clear, as references/sources:Exc. that the hardware speeds do not make so BIG role: ResultsPlus in this test: not so much Elo gain 3 Cores vs 1 Core: ResultsFor anyone missed, as NON-NNUE, Elo diff.is HUGE: Results
And as usual,Many thanks for your interest...
Greetings,Sedat