Hello Ben:
Tennison wrote:Timing = 3 min + 1" / move
Average game length = 7 min 35"
Hash = 512 MB
Book = 8moves.epd
Core(s) = 1 / engine
Code: Select all
1. Stockfish DD x64 (modern) 184.0/300 98-30-172 (L: m=1 t=0 i=0 a=29) (D: r=129 i=30 f=6 s=1 a=6) (tpm=3278.9 d=24.20 nps=1196001)
2. Stockfish 4.0 x64 (sse42-modern) 172.5/300 89-44-167 (L: m=2 t=0 i=0 a=42) (D: r=119 i=28 f=9 s=1 a=10) (tpm=3380.1 d=22.53 nps=1172823)
3. Stockfish 3.0 x64 (popcnt-ja) 122.5/300 35-90-175 (L: m=8 t=0 i=0 a=82) (D: r=129 i=27 f=8 s=0 a=11) (tpm=3454.4 d=22.02 nps=1068707)
4. Stockfish 2.3.1 x64 (popcnt-ja) 121.0/300 32-90-178 (L: m=10 t=0 i=0 a=80) (D: r=137 i=29 f=3 s=0 a=9) (tpm=3373.3 d=21.85 nps=1279226)
Detailed results :
Code: Select all
Stockfish DD - Stockfish 4.0 : 52.5 - 47.5
Stockfish DD - Stockfish 3.0 : 69.0 - 31.0
Stockfish DD - Stockfish 2.3.1 : 62.5 - 37.5
Stockfish 4.0 - Stockfish 3.0 : 61.5 - 38.5
Stockfish 4.0 - Stockfish 2.3.1 : 63.5 - 36.5
Stockfish 3.0 - Stockfish 2.3.1 : 53.0 - 47.0
GAMES
I can not download the games because I am not a registered user of ImmortalChess Forum. Anyway, I post here a very simple rating list calculated with my own rating programme (not error bars, not prior, not drawelo model, etc.):
Code: Select all
Round Robin with 4 engines and 300 games per engine.
Total number of games: 600 games.
Engines: Performance: Score:
Engine 01: 3098.79 61.33 %
Engine 02: 3080.36 57.50 %
Engine 03: 3002.41 40.83 %
Engine 04: 3000.00 40.33 %
Mean of ratings: 3045.39 Elo.
I set SF 2.3.1 rating to 3000. It is just a random choice.
Code: Select all
SF DD 3099
SF 4 3080
SF 3 3002
SF 2.3.1 3000
For error bars with 95% confidence, I propose: ± 25 Elo for SF DD; ± 26 Elo for SF 4; ± 25 Elo for SF 3; and ± 25 Elo for SF 2.3.1, not being strict with decimals. If it is accepted, the ratings will be among: [3074, 3124] for SF DD; [3054, 3106] for SF 4; [2977, 3027] for SF 3; and [2975, 3025] for SF 2.3.1. It would be good to know if EloSTAT gives similar results (I say EloSTAT because I often get similar ratings than EloSTAT; of course I wait for Ordo ratings and BayesElo ratings).
Thank you very much for your test!
Regards from Spain.
Ajedrecista.