Stefan Pohl SPCC
Latest Website-News (2017/09/25): Testrun of asmBrainFish 170825 finished. asmBrainFish is asmFish 170825 using a polyglot-version of the latest Cerebellum-Library (not generated automatically by asmFish, but manually by Thomas Zipproth himself, to avoid move-loops and draws out of the Library (the polyglot-version, asmFish generates automatically, does not prevent these problems and plays weaker!)). Download Cerebellum in BrainFish-format and Polyglot-format: here
2017/06/03: Testrun of Stockfish 170526 finished. Now +30 Elo to Stockfish 8 - nice. Next testrun: asmFish 170522. Result not before Friday.
2017/05/18: Testrun of asmFish 170502 finished. No progress, but no measureable regression, too. So, the result of Stockfish 170503 was perhaps an "statistical accident". Time will tell. Next testrun is Komodo 11 (will start on Monday, when Komodo 11 is released). And after that testrun, the next Stockfish testrun will follow. The huge opening-book testrun is still running - result not before Saturday.
2017/05/05: Testrun of asmFish 170426 finished. Stockfish 170423 made a big step forward (+10 Elo)... To be sure about that impressive progress, the testrun of asmFish 170426 was started (including the same patches as Stockfish 170423). Now, we see, that asmFish 170426 made a big progress, too (+12 Elo to asmFish 170328). So, both results are very similar and we can be sure, that the latest patches in the Stockfish-code are a big step forward! The next testrun of Stockfish will start, when a fast compile of a May 2017-Stockfish is online.
2017/04/03: Testrun of asmFish 170328 finished. Finally, there seems to be a small progress. But to be sure about that, we have to wait for the next Stockfish testrun. Testrun of Stockfish 170402 started. Result not before Sunday.
2017/03/12: Testrun of Stockfishish 170305 finished. Next testrun: asmFish 170310. Result not before Sunday.
2017/03/07: Testrun of asmFish 170301 finished. Next testrun Stockfish 170305. Result not before Monday.
2017/03/02: Testrun of Stockfish 170223 finished. No progress (for more than 2 months, now...). Next testrun asmFish 170301. Result not before Wednesday.
2017/02/25: Testrun of Stockfish 170214 finished. No progress. Next testrun Stockfish 170223. Result not before Friday.
2017/02/19: Testrun of asmFish 170211 finished. No progress, again. Next testrun: Stockfish 170214. Result not before Saturday.
2017/02/14: Testrun of Roc 0.2 aborted - more than 4000 games played, no measureable progress to Gull 3 (+3 Elo). Next testrun: asmFish 170211 (asmFish has all new patches included, the Stockfish ultimaiq-compiles are still from 170205...). Result not before Saturday/Sunday.
2017/02/09: Testrun of asmFish 170202 finished. No progress. Next testrun: Roc 0.2. Result not before Wednesday. If Roc 0.2 is running stable and if it is measureable stronger than Gull 3 (Roc is derivative of Gull 3), then Roc 0.2 will replace Gull 3 as an opponent for my Stockfish- and asmFish-testruns.
2017/02/04: Testrun of Stockfish 170129 finished. No progress, again. Disappointing. The 2 new functional patches had no measureable Elo-effect against the 7 strong non-Stockfish opponent engines. Next testrun: asmFish 170202. Result not before Friday.
2017/01/29: Testrun of asmFish 170122 finished. No progress. Testrun of Stockfish 170129 started (2 new functional patches ("Candidate passed pawns" and "Penalty for a quiet ttMove that fails low" (which scored very well in the Framework)) and 2 simplification patches. Result not before Saturday. The patches are applied in asmFish, too (very fast work!!!). So, the next testrun of asmFish will follow immediately, when the Stockfish-testrun is finished.
Long thinking-time tournament updated. 1034 games played (running more than one month, now). asmFish dominates this tournament of the "Big 3". +73 Elo to Houdini 5 and +84 Elo to Komodo 10.3. Outstanding performance of asmFish !!!
Playing conditions:
Hardware: i7-6700HQ (2.6 GHz) Notebook (Quadcore), Windows 10 64bit, 12GB RAM Fritzmark: 4 cores / 7 threads: 18.4 / 8825 (CPU-speed reduced to 99% (to switch off the Intel Turbo Boost)) Hash: 4 GB per engine GUI: Fritz 13 GUI (no draw, no resign) Tablebases: None for the engines, 3-5 Nalimov for GUI Openings: My new 10 moves SALC V2.0 opening-book. Download it at the Downloads & Links section. Ponder, Large Memory Pages: Off Thinking time: 20'+5'' per game/engine (average game-duration: 52', so around 30 games per day are played). Average thinking time in the middlegame is around 40'' per move. asmFish calculates around 7.2 meganodes per second in early midgame (all engines running with 7 hyperthreading-threads, when thinking), Houdini 5 calculates around 7.5 meganodes per second in early midgame. So the played games are high-quality-chess.
Round Robin (asmFish 161217, Komodo 10.3, Houdini 5)
(Ordo-calculation fixed to Houdini 5 = 3350 Elo)
Program Elo + - Games Score Av.Op. Draws
1 asmFishW_2016-12-17_bmi2 : 3423 12 12 690 60.9 % 3345 57.2 %
2 Houdini 5 Pro x64-pext : 3350 12 12 689 45.6 % 3381 61.1 %
3 Komodo 10.3 64-bit : 3339 12 12 689 43.4 % 3386 56.6 %
2017/01/22: Testrun Stockfish 170113 finished. No progress, again. Next testrun: asmFish 170122. Result not before Saturday. Long thinking-time tournament updated. 868 games played after restart. Stay tuned.
2016/12/14: Testrun of asmFish 161207 finished. The asmFish-project is no longer done by Mohammed Li. But the new version runs incredible fast (+20% faster in singlecore-mode compared to Stockfish (ultimaiq-compiles) (Mohammed Li's versions were "only" +13% faster on my machines) and stable. New asmFish played 7000 games - no crashes, no timelosses - perfect! Next testrun: Stockfish 161212. Result not before Monday.
Program Elo + - Games Score Av.Op. Draws
1 BrainFish 161009 numa : 3467 8 8 7000 85.0 % 3151 24.7 %
2 BrainFish 161128 x64 : 3464 8 8 7000 81.3 % 3188 30.5 %
3 asmFish 161207 x64 : 3427 7 7 7000 78.1 % 3188 33.6 %
4 asmFish 161217 x64 : 3426 7 7 7000 78.0 % 3188 32.9 %
5 asmFish 170109 x64 : 3426 7 7 7000 76.8 % 3201 33.9 %
6 asmFish 161004 x64 : 3421 8 8 7000 81.5 % 3151 29.2 %
7 Stockfish 161212 x64 : 3408 7 7 7000 76.4 % 3188 35.8 %
8 Stockfish 170105 x64 : 3404 7 7 7000 74.7 % 3201 36.0 %
9 Stockfish 170113 x64 : 3403 7 7 7000 74.6 % 3201 35.7 % (new)
10 Stockfish 161120 x64 : 3395 7 7 7000 75.1 % 3188 35.7 %
11 Stockfish 161127 x64 : 3394 7 7 7000 74.9 % 3188 36.2 %
12 Stockfish 8 161101 : 3390 5 5 11000 74.3 % 3191 36.1 %
13 Houdini 5 x64 : 3368 4 4 18000 56.8 % 3310 45.6 %
14 Komodo 10.3 x64 : 3339 5 5 11000 60.5 % 3253 37.9 %
15 Komodo 10.1 x64 : 3322 6 6 8000 65.0 % 3203 33.3 %
16 Komodo 10.2 x64 : 3315 4 4 15000 54.2 % 3279 38.8 %
17 Houdini 4 x64 : 3194 5 5 11000 44.7 % 3240 34.7 %
18 Shredder 13 x64 : 3184 4 4 19000 33.8 % 3314 35.9 %
19 Fizbo 1.9 x64 : 3164 5 5 10000 34.8 % 3287 30.1 %
20 Gull 3 x64 : 3127 4 4 23000 27.2 % 3318 33.2 %
21 Fire 4 x64 : 3116 4 4 23000 26.0 % 3319 33.1 %
22 Critter 1.6a x64 : 3112 4 4 23000 25.6 % 3319 30.5 %
23 Equinox 3.3 x64 : 3094 4 4 19000 24.5 % 3314 31.7 %
24 Mars 3.41 x64 : 3092 6 6 10000 30.7 % 3256 34.5 %
The Elo-ratings of older Stockfish dev-versions in the Ordo-calculation can be a little different to the Elo-"dots" in the diagram, because the results/games of new Stockfish dev-versions - when getting part of the Ordo-calculation - can change the Elo-ratings of the opponent engines and that can change the Elo-ratings of older Stockfish dev-versions (in the Ordo-calculation / ratinglist, but not in the diagram, where all Elo-"dots" are the rating of one Stockfish dev-version at the moment, when the testrun of that Stockfish dev-version was finished).
Individual statistics:
asmFish 161207 x64 : 3427 7000 (+4292,=2350,-358), 78.1 %
Houdini 5 x64 : 1000 (+327,=535,-138), 59.5 %
Shredder 13 x64 : 1000 (+640,=324,- 36), 80.2 %
Komodo 10.2 x64 : 1000 (+446,=459,- 95), 67.5 %
Critter 1.6a x64 : 1000 (+729,=243,- 28), 85.0 %
Gull 3 x64 : 1000 (+694,=279,- 27), 83.3 %
Fire 4 x64 : 1000 (+715,=266,- 19), 84.8 %
Equinox 3.3 x64 : 1000 (+741,=244,- 15), 86.3 %