Stefan Pohl Computer ChessHome of famous UHO openings and EAS RatinglistSPCC Super 3 Tournament
Tournament updates are suspended at the moment, because I want my new project Cerberus participating here. Because the Cerberus project is still in development and alpha-testing, updates of the Super 3 Tournament are not possible at the moment. Stay tuned for more information to come in early 2025...
Latest update: 2024/11/25 (next updates will follow every 10-12 days...) Tournament overall runtime (average: 47.2 games per day are played): 193 days Download all played games since start (May 2024) here
An endless RoundRobin-tournament with 3 engines, which are at the same level of strength (around 3400 Elo) but are completely different in their inner structure and their way of thinking. Why? The strongest engine since more a decade (Stockfish) is open source, so many, many other engines are (at least) "inspired" by Stockfish... And additionally, a lot of engines (including Stockfish!) are using Lc0-training-data for building their neural nets: The high-end computerchess has become very incestuous... To say this clear here: This is not good or bad, it is just the reality of high-end computerchess in these days. So, IMHO, it is very interesting to run a tournament with engines, which are completely different, not only in their playing-style, but also in their inner structure and way of thinking, but on a close level of playing-strength. The Super 3 tournament is not about the results (as you can see below, all 3 engines are at the same level of strength), but about generating interesting enginegames. | Search | Evaluation | nodes per second (early middlegame) As you can see, these 3 engines have nothing in common, considering their way of thinking. Komodo 14.1 uses a classical handcrafted evaluation, Lc0 CPU uses a (float) neural net, and Revenge 1 uses a (integer) nnue-net, like most modern engines in these days. Because floating-point calculations are brutally slow on CPUs, Lc0 CPU is way slower than the 2 opponents... (this is the reason, why Lc0 normally runs on the GPU, not the CPU). If you want to learn more about the neural-net of Lc0 and about nnue-nets, I recommend this e-book by Dominik Klein, which can be downloaded for free as a PDF-file. And, additionally, Lc0 uses a complete different search (MCTS). And Revenge 1 is one of the most aggressive playing engines of all time (see the EAS-Ratinglist below).
Hardware: AMD Ryzen 7840HS 8-core (16 threads) notebook with 32GB RAM. Turboboost off. Speed: See above. Each engine uses 14 threads, when thinking (Lc0 cpu dnll has the UCI option "Threads" like any normal CPU-engine, so it uses the CPU like all other engines) - the GPU stays (of course) unused. Hash: 8 GB per engine (20.000.000 NNCachesize for Lc0 - enough for storing all evaluated positions of a complete game) GUI: CutechessGUI (GUI ends game, when a 6-piece endgame is on the board, all other games are played until mate or draw by chess-rules (3fold, 50-moves, stalemate)) Tablebases: None for engines, 6 Syzygy for CutechessGUI Openings: My UHO_2024_8mvs_+085_+094.pgn openings are used (randomly mixed, each opening repeated with reversed colors, of course (=Gamepairs)) Ponder, Large Memory Pages & learning: Off Thinking time: 10min+5sec per game/engine (average game-duration: 30 minutes), so only 50 games are played in 24 hours = high quality enginechess
Here you can see the shortest wins with sacrifices, played between the latest two site-updates, filtered by my Interesting Wins Search Tool. Download this cool tool in the "Downloads & Links" section or right here Many thanks to ChessBase for the pgn-replayer tool, which is very easy to use (only 3 lines of code!) and very powerful - use the fan (propeller?)-icon right near the arrows below the chessboard, to start and stop the online-analyzing with the Fritz-engine! Perhaps you have to clear your browser-cache to see the latest games - otherwise the pgn-replayer does not update the games correctly...if you can not see the chessboard, check, that your browser has Javascript activated or if an AdBlocker is the problem.
Below the results (first normal Celos (by ORDO), followed by gamepair-rescored Celos, followed by EAS-Ratinglist).
Program Celo + - Games Score Av.Op. Draws 1 Lc0 791921 CPU : 3450 6 6 5738 51.7% 3438 49.0%
Games : 9096 (finished) White Wins : 4376 (48.1 %) Gamepairs: # PLAYER : Celo Error Pairs W D L (%) CFS(%)
-------------------------------------------------------------------
Head to head statistics:
1) Rebel Extreme 1.0 3450 : 489 (+135,=241,-113), 52.2 % vs. : pairs ( +, =, -), (%) : Diff, SD, CFS (%)
2) Lc0 791921 CPU 3450 : 2869 (+802,=1443,-624), 53.1 % vs. : pairs ( +, =, -), (%) : Diff, SD, CFS (%)
3) Komodo 14.1 HCE 3431 : 2869 (+712,=1397,-760), 49.2 % vs. : pairs ( +, =, -), (%) : Diff, SD, CFS (%)
4) Revenge 1.0 avx2 3423 : 2869 (+649,=1419,-801), 47.4 % vs. : pairs ( +, =, -), (%) : Diff, SD, CFS (%)
Here the EAS-Ratinglist, calculated by my EAS-Tool: bad avg.win
A: Most high-value sacrifices (3+ pawnunits) : [1]:05.32% Revenge 1.0 avx2
|