NNUE Strength Testings (played mainly for XLVII tour):
GAMES
Both engines are BMI2, played with 1 Core and 128 MB hashAnd both engines are used to play via nn-c3ca321c51c9.nnueOverall draw percentage (based on 1258 games) is high 95 %For more details about the current strength match: NNUE DuelThe Elo Difference is 5--------------------------------------------------------------------------------------------------2nd Test1 SF Polyglot 211120 1Core +84/-15/=239 60.21% 203.5/3382 SF200620MZ C40 3 Cores +15/-84/=239 39.79% 134.5/338
SF Polyglot used to play via nn-c3ca321c51c9 and 128 MB hashSF200620MZ is played with Contempt 40, 3 Cores, 512 MB hashThe Elo Difference is 72--------------------------------------------------------------------------------------------------3rd Test: CF EXT with 256 MB Hash1 CF EXT 161120 1 Core +25/-6/=83 58.33% 66.5/1142 SF200620MZ 3 Cores +6/-25/=83 41.67% 47.5/114
CF EXT 161120 used to play via nn-cb26f10b1fd9 and 256 MB hashSF200620MZ is played with Contempt 40, 3 Cores, 512 MB hashThe Elo Difference is 58--------------------------------------------------------------------------------------------------4th Test: CF EXT with 128 MB Hash1 CF EXT 161120 1 Core +127/-9/=334 62.55% 294.0/4702 SF200620MZ 3 Cores +9/-127/=334 37.45% 176.0/470
CF EXT 161120 used to play via nn-cb26f10b1fd9 and 128 MB hashSF200620MZ is played with Contempt 40, 3 Cores, 512 MB hashThe Elo Difference is 89--------------------------------------------------------------------------------------------------5th Test: 1 Core vs 3 Cores1 CF EXT 161120 3 Cores +31/-1/=328 54.17% 195.0/3602 CF EXT 161120 1 Core +1/-31/=328 45.83% 165.0/360
Both participants are used to play via nn-cb26f10b1fd93 Cores played with 512 MB hash / 1 Core with 128 MB hashThe Elo Difference is 29
GAMES (2nd, 3rd, 3th, 4rd, 5th tests)--------------------------------------------------------------------------------------------------More Details:Perfect 2020 opening book is used for all strength engine testingsCF EXT 161120 seems to be the strongest under these conditionsNote also that the same CF managed to be Champ of NNUE IIWe noticed that 128 MB hash performed better than 256 MB hashIt seems (as NNUE), 3 cores do not gain too much over 1 coreE.g, approx. 30 Elo, but as NON-NNUE, the Elo difference 140Note that I am referring about the played results: Deep BrainWhere in the recent strength engine testings, we noticed also:NNUE 1 core is approx. 90 Elo stronger than NON-NNUE 3 coresJust impressive, this is the main reason about why I preferred...!That means also, XLVII is stronger than previous book tourneys!Btw, why I preferred also 1 Core and 3 Min (about XLVII tour)Otherwise...in case of giving more Time and plus more Cores...Then probably the draw perc. is expecting to be almost 98-99%So, with my preferred conditions, I expect to see less draws....But this is also true, via NNUE too many draws...and too boring...Normally I like draws as well, but enough is as good as a feast!)And as I stated earlier, I believe in that:Via NNUE, it's expecting to see stronger/better quality games!But NNUE testings are not ideal way for checking the openingsYes, XLVII's current draws are 93%, but they came mainly fromVery critical/ weaker books, but in case of strong book matchesThen the results are different, almost all are ending as draws...Anyhow sometimes, we have to do something different...right ?)Sure, in case of next tours, I hope to switch again to NON-NNUE
Thanks for reading...
Best Regards,Sedat
Standings / More Details