The Week in Chess

Tuesday, April 23, 2013

Stockfish 130419 Defeated Houdini 3 Pro!

The stage 4 live tournament has just completed. Houdini 3 won that stage over Stockfish, Rybka and Komodo.  The great superfinal match between Houdini and Stockfish is now set at 48 games.  My previous post pointed out that the finals will be between these two chess engine titans (which became true) and the eventual winner will be Houdini.

I watched some of the games of the 4th stage and observed that the fight were so close, with so many draws similar to the top human chess grandmasters tournament. Stockfish was the leader for the short time until the last few rounds when Houdini took control. The games between Houdini and Stockfish seemed a do or die situation where it was the only match between two engines that there were more wins and losses than draws. Houdini has the fastest processing speed in terms of nodes per second but could not score a single win against Rybka with the slowest speed. Even Stockfish could only manage single wins agains Rybka and Komodo. The surprise of the live tournament was the luck of Komodo to reach stage 4 with only single core power, but even with 15x less power, it scored pretty well against the multi-core engines who were using the full 16 cores of TCEC's monster computer. If Don Dailey could finally complete his overdue ongoing multi-core engine development, it is an engine to watch.

With this observation, it is possible that Houdini could be beaten by Stockfish and Rybka given a few tweaks. But Rybka is seemingly abandoned by Vas Rajlich, so it is Stockfish that has the best chance to beat Houdini  especially that it is allowed by TCEC to change engines in the course of the tournament, and Stockfish has already used this substitution feature twice.  The Stockfish github was frozen while the stage 4 was in progress, probably the team is cooking a fine tuned Stockfish to be served at the superfinal.

While stage 4 was in progress, I made my own tests whether Houdini could be beaten by Stockfish by trying different combinations of compiler options and removing some "baggage" in the Stockfish source code. I am like a monkey assembling a wristwatch from discrete components because I am not an expert in C++, (I am still trying to understand what are templates, namespaces, overloading and other complicated C++ stuff. What I know is Foxpro!)  The compiler used was Visual Studio 2010 and the source was the last build on April 19.

Anyway, my tests produced a surprising result.  My Stockfish compilation demolished both Houdini 3 with 6 wins lead and the latest Stockfish  auto-build dated April 19, 2013 with 8 wins margin. Take this with a grain of salt and those interested may try my compilation with their computers, because my experience in rating list production tells me not to trust a single result as absolute since statistics are random that I have no control over.

Here is the result.

Stockfish 130419 Tests
1Stockfish 130419 x64 msc pgo107.0/200· ·· ·· ··
2Houdini 3 Pro x64 47.0/10032-38-30
3Stockfish 13041910 x64 46.0/10020-28-52

200 games played / Tournament finished

Tournament start: 2013.04.22, 15:17:25
Latest update: 2013.04.23, 15:19:54
Level: Blitz 0:10/1
Hardware: AMD Phenom(tm) II X4 945 Processor 3000 MHz with 4 GB Memory
Operating system: Microsoft Windows XP 64 Bit Professional Service Pack 2 (Build 3790)
Table created with: Arena 3.0

Download Stockfish 130419 x64 msc pgo here.

Download computer chess engines tournament games here.

No comments:

Post a Comment

Chessdom News