Not logged inCSS-Forum
Forum CSS-Online Help Search Login
CSS-Shop Impressum Datenschutz
Up Topic Hauptforen / CSS-Forum / Neues SSDF Ranking erschienen
- - By Lothar Jung Date 2023-01-20 11:27 Edited 2023-01-20 11:30
https://ssdf.bosjo.net/list.htm

An der Spitze geht es sehr eng zu.

Lc0 hauchdünn Nr. 1.

Hier die Anmerkungen von Jens Sandin:

Comments

to the Swedish Rating List 3/2022, December 31

In this last rating list of the year 2022, we can present five new entries. First one out is the latest version from team Stockfish, namely Stockfish 15. This version landed just a hairline from the number one spot this time and grabbed the third position, but only one single point behind the leader Lc0! After 480 played games it has earned a rating of 3574, which is 17 points ahead of Stockfish 14 and exactly on par with Stockfish 13 at the moment. As opening book for Stockfish 14, we have used Fauzi Dabat's "Aggressive 5.4 by Fauzi.abk".

Next one out is Dragon Komodo 3 by Mark Lefler and Larry Kaufman. We have tested both the "normal" version and the Monte Carlo Tree Search-version. The non-MCTS version of Dragon Komodo 3 earned a fifth place in this rating list, with a rating of 3571 after 285 games played. It has positioned itself five points ahead of the 2.51 version and just four points behind the number one spot at the moment!

In the testing of the Monte Carlo Tree Search (MCTS) Dragon Komodo, we have used the 3.1 version. It has recieved a rating of 3519 after 306 games played. It is 41 points stronger than the Dragon Komodo 2 MCTS and took the eleventh spot on this rating list. As opening book for both the ordinary and the MCTS-version, we have used Erdogan Günes opening book "out10-35.bin". The sharp-sighted has probably noticed that the nomenclature "x64" is omitted from the name this time. We can insure that the testing has been done with the strongest settings, AVX2 and also 64-bit, and that the omission is made just because of the length limitations in the naming of the engines.

I can also add that since most engines that we have tested on both the Q6600 and 1800X hardware has been 64-bit programs, we will soon make an overhaul to upcoming rating lists and strip this rather unnecessary designation from the names - only adding 32 bit to the names of the few engines that hasn't been 64-bit compatible.

Well, to get back on track. The next program that we can present in our latest rating list, is Alex Morozov's - Booot 7. This time he has used his own implementation of Neural Net technology (NN) in evaluation, using Python, Delphi and SIMD. This has proved to been useful to the already strong Booot- program and he has managed to get a rating of 3471 after the first 185 played games. This is 42 points ahead of the "ordinary evaluation"-version of Booot 6.5 and it now holds the first spot of the engines outside the three in the top: Lc0, Stockfish and Dragon Komodo. As opening book for Booot 7, we have used the "Perfect2021.abk" opening book by Sedat Canbaz.

Last one out in this rating list is Martin Danielsson with his Marvin-program. We have tested the 6.0.0-version this time, and it has recieved a rating of 3365 after 220 played games. It is 37 points stronger than the last Marvin version (5.2.0) which we have tested and we have used Marvin's own opening book for the testing.

Since the last rating list, Wasp 4.5 1800X gained 14 points, Pedone 3.1 1800X gained 10 and Arasan 23.01 1800X gained 9 points. Biggest loss since the last list was Lc0 0.28.2 which dropped 16 points in the lead. Other notable losses was Roma II which lost 13 points and Chiron 5.01 1800X, which lost 8 points.

The next rating list will probably be released relatively soon and will feature the manually tested Revelation II A.E. Komodo 12.1, and hopefully also contain some more engines on both the 1800X and Q6600 hardware level. We will also begin testing of a more recent version of Lc0 soon, so hopefully it will have enough games to be presented as an early bird in the next rating list.

Lars Sandin
Parent - By Reinhold Stibi Date 2023-01-20 11:48
So einen Blödsinn.

Da kann sich jetzt Lc0 gegen Stockfish 13 sonnen; total veraltet inzwischen gibt es
Stockfish 15.1

 
Parent - - By Walter Eigenmann Date 2023-01-20 12:22
.

Lustig ist ja (u.a.), dass dort Stockfish 13 und Stockfish 15 exakt "gleich stark" sind:

Code:
                                        Rating  +     -    Games   Won   Av.opp
01 Lc0 0.28.2 Cuda-611213 3060Ti        3575    27   -25   772     66%   3447
02 Stockfish 13 x64 1800X 3.6 GHz       3574    35   -32   480     69%   3436
03 Stockfish 15 x64 1800X 3.6 GHz       3574    42   -39   320     66%   3458
04 Lc0 0.26.3 Cuda(67362) 3060Ti        3571    29   -27   680     68%   3444
05 Dragon Komodo 3 x64 1800X 3.6 GHz    3571    43   -41   285     62%   3487
06 Dragon Komodo 2.51 x64 1800X 3.6 GHz 3566    42   -39   320     65%   3457
07 Stockfish 14 x64 1800X 3.6 GHz       3557    39   -37   360     66%   3445
08 Stockfish 12 NNUE x64 1800X 3.6 GHz  3554    28   -27   658     61%   3464
09 Dragon by Komodo x64 1800X 3.6 GHz   3542    34   -32   460     63%   3448
10 Stockfish 11 x64 1800X 3.6 GHz       3535    36   -34   450     70%   3391

Käme dieses Ranking von einem Stellungstest, wäre das Geschrei groß...
(Sorry, konnte ich mir nicht verkneifen...

Aber ok, ist ja nicht auszuschließen, dass auf Turnierstufe ("40 moves/2 hours followed by 20 moves/each following hour")
die Relationen anders als gewohnt sind.

Gruss: Walter

.
Parent - By Lothar Jung Date 2023-01-20 12:29
Ja Walter,

bei Turnierbedingungen verkürzen sich die Abstände erheblich.

Der Schachgott setzt den Hobel an und hobelt alle gleich.

Gruß

Lothar
Parent - - By Benno Hartwig Date 2023-01-20 14:59 Upvotes 2

> Käme dieses Ranking von einem Stellungstest, wäre das Geschrei groß...


Bei SSDF schreit man eben nicht, man lächelt nur.
Und man lächelt wohlwollend, wenn man sich eben an die SSDF-Verdienste in der Computerschach-Frühzeit erinnert.
SSDF waren soweit ich weiß eben die ersten, die eine echte Rankingliste erstellten und regelmäßig weiterpflegten, und dies mit unendlich viel händischem Fleiß, als tapfer immer wieder die Figuren von Brettcomputern zu setzen waren.
Parent - By Lothar Jung Date 2023-01-20 15:44 Upvotes 1
Gerade für Brettcomputer und ältere Engines sollte man der SSDF Respekt zollen.
Parent - - By Walter Eigenmann Date 2023-01-20 16:19
.

Selbstverständlich, die SSDF-Tester haben große Verdienste, und trotz der  bescheidenen Hardware der Schweden
waren ihre Games mit Turnier-Bedenkzeiten eine wichtige Quelle des Partienstudiums, nicht zuletzt damals in der Schachcomputer-Szene.

BTW: Ist jemandem schon gelungen, die betr. PGN's runterzuladen? Ich kriege hier nur immer eine "Zeitüberschreitung"...

.
Parent - By Reinhold Stibi Date 2023-01-20 17:16
Ja,, die SSDF-Tester hatten früher große Verdienste; nur in der Gegenwart nicht.

Wenn die von der Spielstärke Stockfish 13 und 15 gleich setzen, dann stimmt doch
da offensichtlich was mit den Testmethoden nicht; z.B. die Wahl der Eröffnungen.
Up Topic Hauptforen / CSS-Forum / Neues SSDF Ranking erschienen

Powered by mwForum 2.29.3 © 1999-2014 Markus Wichitill