Deep Rybka 5

rybka is the best, and only, current engine which can be used for play. it has a lot of good features, and its usefulness has been proven. but rybka is a commercial product. as a result, i will make a prototype of rybka, and create a new engine which has some competitive features for free. moreover, i plan to test which ones are useful in the end, and gradually add these features to rybka. then i will release the engine to the public for free. my intention is to make a fruit engine that is more complex than rybka, but simpler than rybka. i will make a new engine that behaves more like rybka, but is simpler in construction. i will compare its effectiveness with that of rybka.

shredder introduces the concept of node. it is a node that represents a group of nodes. for example, you can analyze a position by searching for a node in a node tree that gives you a promising node. or you can search a node tree that gives you a promising node. the search is taken into consideration to determine the node which gives the most promising node.

houdini is a hybrid engine which in search strategy goes for a conservative approach, pruning its tree of analysis early. it is inordinately fond of drawing by sacrificing material, and while fischer randomisation can help to compensate, houdini lacks something, compared to komodo, which has sacrificed less.

another one of the surprises of rybka is the warmth of its „tactics.” it has a surprising ability to go for tactical sacrifices which is in line with its ultra-aggressive nature. it is also strong in the endgame. these things are still in evidence, despite the perceptive analysis by daniel king. there seems to be a lot of material left for the author of the report, a programmed search for the perfect completion to take advantage of. it is indeed hard to compare the strength of the various engines in tcec tiebreaks which are played very fast, while rybka is likely to have an advantage over houdini or stockfish, which are rather slower.

gm grigory rybka is a world-class grandmaster, former ussr champion, and long-time coach at the lugansk state chess institute. rybka uses rybka 5 from the chessbase distribution and has been part of the chessbase app since version 0.9.1. test #3: i decided to run about 50 problems from the wccc 2010 and wccc 2011. the wccc 2010 problems go up to around 1800, and the wccc 2011 problems go up to around 2000, although most of them are around 1700. i was using computers from my collection, including a macpro 6, 2 x 2.8 ghz, 6 gb ram and two dual cpu systems, a bluegene/q (supercomputer) and a macpro 6, 2 x 2.8 ghz, 6 gb ram and 8 xeon cpus. these test cases were tested in two different ways: i tested houdini 4 in both alpha and beta mode. i also tested the two komodos in alpha and beta mode, as well as stockfish 5 alpha, stockfish 5 beta, deep rybka 4.1 in alpha and beta mode. i compared these to stockfish 5, the komodos and rybka 4.1 in their latest versions. i also compared rybka 4.1 alpha to rybka 4.1 beta and rybka 4.1 to rybka 4.1 alpha. test #4: for this test i selected about 75 of the first 50 problems from the wccc 2011. i ran about 50 of the wccc 2011 problems using 4 computers, as well as having a couple of other systems on hand to check. these tests were done on the bluegene/q with two computers: a macpro6, 2.8 ghz, 12 gb ram, 3 x intel cores, six $$$$$ graphics cards, apple xserve g5 = 1tb hdd, three ethernet connections, an internet connection, and tons of hard drive space. i tested rybka on windows vista, rybka 4.1 alpha, rybka 4.1 beta and rybka 4.1. i compared this to the komodo’s 6.4 alpha and 6.4 beta. 5ec8ef588b