Rybka Chess Community Forum
I'll post the 2nd game later in the week, after I play it.
2 Games between Deep Rybka 4 SSE42 x64 and Naum 4.2
Conditions G 12 Hours + 30 Seconds Per move - 1 Gig Of Hash Each
Full set of 5 Man Tablebases
I7-920 2.67 GHZ
Gui - Aquarium 4.05 Build 236
rnbqkb1r/ppp2ppp/4pn2/4P3/2Pp4/2N5/PP1P1PPP/R1BQKBNR w KQkq - 0 1
Rybka won the first game with Black. For those of you criticizing Rybka endgame play, I don't see much wrong with the way she handled the Queen and Pawn Endgame. In my tests from this position, with this time control, Rybka is 3-0. See my other thread for the 2-0 result against Zappa Mexico II - http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=271025;hl=
12 hours + 30 seconds!
I thought this time control would be a good way to get to the truth about engine strength.
No problem, I misunderstood your post the first time. I got the impression that it was 12 hours and 30 seconds per move.
12 hours per move would be a very long wait for the game to finish. It would be nice if you do that on a spare PC.
I wish I had a spare pc just to run these tests on. I don't mind trying up the pc once in a while, as I get to prove Rybka's superiority. After this match, I'm probably going to repeat the test, against stockfish 1.8 and critter 0.8.0. Then, I'll set up some Kings Indian Positions and test,using the same time controls. My tests showed that Rybka 3 was a little better than stockfish 1.7, however, in the following position, stockfish played better, from both sides. I don't remember the score. I'll check later and post it. I'm curious how Rybka 4 handles the following position (both sides) against stockfish 1.8, with a long time control:
rnbq1rk1/pp3pbp/3p2p1/3Pp2n/4P3/2NBBP2/PP4PP/R2QK1NR w KQ - 0 10
> I thought this time control would be a good way to get to the truth about engine strength.
No, to get the engine's true strength you need enough games (and well, if you get enough games at that time control, that's fine), otherwise you're just playing 1' + 3" bullet time controls from year 2024.
Your comment about future bullet games reminds me of a conversation that I had with Larry Kaufmann, about long time controls (40/29700+90:20/3600+90:1800+90) that I was running between Rybka 3 and Deep Sjeng 3.0. As the games were being played on my old single core P4, Larry Commented that these were the equivalent of fast time controls (I don't remember and can't find his exact comment, he may have said bullet) on more modern multi core pc's. I wonder how accurate your statement is. Maybe the games that I am playing will be the equivalent of playing 1' + 3" bullet time controls in the year 2018?! Can this be mathematically determined? Hardware progression and software advances are freaky. Om my old 386, at one time, the latest version of M-Chess was the best software, and my favorite. I strongly suspect that If I were to run a 1,000 game handicap match between the engines, with Rybka 4 on my Quad Core, taking 5 seconds per move, and M-Chess on a 386, taking 1 month per move, Rybka would win 1,000 to 0.
I think Rybka 4 would win that match with zero calculation, using nothing except her static positional evaluation.
> Maybe the games that I am playing will be the equivalent of playing 1' + 3" bullet time controls in the year 2018?!
I just went with "doubling of power every 18 months" or Moore's Law, so that in 2018, it would take a 20 minutes + 3 seconds time control to reproduce your results of 12 hours + 30 seconds time controls. Unless there's stalling in progress or a breakthrough (like the technological singularity, though that's predicted till 2080 or so).
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill