Discussion of computer chess matches and engine tournaments.
Moderators: hgm , Rebel , chrisw
Sylwy
Posts: 4465Joined: Fri Apr 21, 2006 4:19 pmLocation: IASI - the historical capital of MOLDOVAFull name: SilvianR
Post by Sylwy Wed Jul 21, 2021 6:48 pm
Test conditions:
-TC=4'+2"
-Hash=256 MB
-GUI: Arena 3.5.1
-Book: own Book.bin-(Pedone 3.1)-(for Revenge_1.0_x64_bmi2) and own TestBook.scb (for SlowChess Blitz 2.6 x64-avx)
-default settings for both engines
-1 thread-CPU=Intel i5-7400-3GHz (Kaby Lake)
-TBS: 6-men Syzygy bases for both engines + EndgameBitbases.sbb+NeuralNets.snn for SlowChess Blitz 2.6 x64-avx
-OS: Windows 10 Home.
+27 -10 =63 +63 Elo points
SlowChess Blitz x2.6 has 3477 Elo points (CCRL Blitz).
Expected rating for Revenge_1.0_x64_bmi2 is around 3540 Elo points (in my single match test).
If RubiChess 2.2 will be defeated (as the beginning of the match seems to look like) then
Revenge is without a doubt the fourth chess engine (1 CPU league) in the world (after SF 14-FF2 & KomodoDragon). An exceptional project by Fabio Gobbato. Congratulations !
Sylwy
Posts: 4465Joined: Fri Apr 21, 2006 4:19 pmLocation: IASI - the historical capital of MOLDOVAFull name: SilvianR
Post by Sylwy Fri Jul 23, 2021 8:31 am
Tonight the final result. Revenge_1.0_x64_bmi2 is a very good NNUE chess engine.
Sylwy
Posts: 4465Joined: Fri Apr 21, 2006 4:19 pmLocation: IASI - the historical capital of MOLDOVAFull name: SilvianR
Post by Sylwy Fri Jul 23, 2021 4:39 pm
Final result in my second test.
-------------------------------------
Test conditions:
-TC=4'+2"
-Hash=256 MB
-GUI: Arena 3.5.1
-Book: own Book.bin-(Pedone 3.1)-(for Revenge_1.0_x64_bmi2) and Performance.bin (for RubiChess-2.2_x86-64-bmi2)
-default settings for both engines
-1 thread-CPU=Intel i5-7400-3GHz (Kaby Lake)
-TBS: 6-men Syzygy bases for both engines
-OS: Windows 10 Home.
+20 -6 =74 +49 Elo points
RubiChess-2.2_x86-64-bmi2 (NN-fb50f) has 3455 Elo points (CCRL Blitz).
Over 3500 ELO points also in my second test. Clearly (for me) Revenge_1.0_x64_bmi2 is currently the fourth chess engine in the world (in the 1CPU league).
==============
Next: Koivisto_6.0-x64-windows-avx2
RubiChess
Posts: 584Joined: Fri Mar 30, 2018 7:20 amFull name: Andreas Matthies
Post by RubiChess Fri Jul 23, 2021 6:26 pm
Sylwy wrote: ↑ Fri Jul 23, 2021 4:39 pm
Test conditions:
...
-Book: own Book.bin-(Pedone 3.1)-(for Revenge_1.0_x64_bmi2) and Performance.bin (for RubiChess-2.2_x86-64-bmi2)
+20 -6 =74 +49 Elo points
RubiChess-2.2_x86-64-bmi2 (NN-fb50f) has 3455 Elo points (CCRL Blitz).
Over 3500 ELO points also in my second test. Clearly (for me) Revenge_1.0_x64_bmi2 is currently the fourth chess engine in the world (in the 1CPU league).
You compare to CCRL Elo but allow using own book (which none of the known ranking lists does)?
We will see what the official rankings will show. Ipmanchess to my knowledge has Revenge 1.0 < Rubi 2.2 so far.
Regards, Andreas
Sylwy
Posts: 4465Joined: Fri Apr 21, 2006 4:19 pmLocation: IASI - the historical capital of MOLDOVAFull name: SilvianR
Post by Sylwy Fri Jul 23, 2021 9:18 pm
RubiChess wrote: ↑ Fri Jul 23, 2021 6:26 pm
You compare to CCRL Elo but allow using own book (which none of the known ranking lists does)?
In general I use the default packages (like was also in the first post-the test for SlowChess Blitz 2.6). RubiChess default packages hasn't a default book. Well, Performance.bin is a good one. And in most games, after the few moves from book, the positions were practically equal. I wonder if Ipmanchess tests on 1CPU, with the similar hardware and CCRL-like test conditions.
Regards,
Silvian
Sylwy
Posts: 4465Joined: Fri Apr 21, 2006 4:19 pmLocation: IASI - the historical capital of MOLDOVAFull name: SilvianR
Post by Sylwy Sun Jul 25, 2021 6:43 pm
A rating (CCRL Blitz) and two tests that seem like a joke to me:
vs.SlowChess Blitz x2.6
+27 -10 =63 ...............my test (100 games)
+13 -16 =69 ...............CCRL test (98 games)
vs.RubiChess 2.2
+20 -6 =74 ...............my test (100 games)
+11 -25 =62 ...............CCRL test (98 games)
Both my tests are above.
Possible differences:
1.- I used bmi2 versions (for Revenge 1.0 and RubiChess 2.2) and avx for SlowChess Blitz 2.6; I don't know which compilations use CCRL;
2. my CPU is Intel i5-7400-3GHz; I don't know on which CPUs the CCRL tests were;
3. I used native books (Revenge and SlowChess Blitz) and Performance.bin for RubiChess. But limited to a small number of moves that did not generate significant differences when leaving the book.
We also can't talk about a significant number of extra games.
Strange results. At least for me who watched most of the 200 games on my PC!
Sylwy
Posts: 4465Joined: Fri Apr 21, 2006 4:19 pmLocation: IASI - the historical capital of MOLDOVAFull name: SilvianR
Post by Sylwy Sun Jul 25, 2021 8:53 pm
Again, in all my tests I always use the author's compilations (default packages). For Revenge 1.0 it's clear - it's closed source. But both SlowChess Blitz 2.6 & RubiChess 2.2 are open sources. CCRL uses in their tests the authors compilations, or .........?
RubiChess
Posts: 584Joined: Fri Mar 30, 2018 7:20 amFull name: Andreas Matthies
Post by RubiChess Sun Jul 25, 2021 8:58 pm
Sylwy wrote: ↑ Sun Jul 25, 2021 8:53 pm
CCRL uses in their tests the authors compilations, or .........?
Most probably they do. If they compile themself and have a faster Windows binary than mine I would be happy if they contact me and show me their compiler settings.
Sylwy
Posts: 4465Joined: Fri Apr 21, 2006 4:19 pmLocation: IASI - the historical capital of MOLDOVAFull name: SilvianR
Post by Sylwy Sun Jul 25, 2021 9:09 pm
RubiChess wrote: ↑ Sun Jul 25, 2021 8:58 pm Sylwy wrote: ↑ Sun Jul 25, 2021 8:53 pm
CCRL uses in their tests the authors compilations, or .........?
........ If they compile themself and have a faster Windows binary than mine I would be happy if they contact me and show me their compiler settings.
I'm sure !
I don't know if the rest of the users who don't benefit from it ...... plus the equity in testing. From what I've heard (I'm not a professional programmer) an individualized compilation on a certain type of computer is the best for that computer. Am I wrong? I would be very curious to hear Fabio's opinion about this question.
