The great 40/120 Tournament

Discussion of computer chess matches and engine tournaments.

Moderators: hgm, Rebel, chrisw

krazyken

Re: The great 40/120 Tournament

Post by krazyken »

Rybka is run 64 bit via microwine.

These are run at 32-bit via wine:
Twisted Logic
Thinker
bright
Spike
Wildcat
Naum
The Baron
Booot
Colossus
ProDeo
Joker
Gibbon
romi

The rest are Mac Native. If anybody knows of any other Mac programs that can be added, please let me know.
Michel
Posts: 2272
Joined: Mon Sep 29, 2008 1:50 am

Re: The great 40/120 Tournament

Post by Michel »

Yes, it has come a long way from it's stable Fruit roots. It appears to be connected to the threading code, as reducing the number of threads increases the time between failure. I will try less aggressive optimizations to see where I get. Have you run many longer time control games with many threaded Toga?
Toga does not lock or otherwise protect its hash table from simultaneous access. So this could lead to problems. I tried to bring this up with the Toga developers at their forum but they did not react. Otherwise I do not really see what could be causing problems. I only did some limited testing on two threads without obvious problems.
P.S. The versions of DLT and Grapefruit on your site do not run at all with multiple threads on My mac
I seem to have used the old semaphore code there. Keeping track of so many Toga version is quite confusing. I will fix it
krazyken

Re: The great 40/120 Tournament

Post by krazyken »

krazyken wrote:Bikjump has some bugs in endgame lookup, getting losses in drawn endings.
It appears there was an error in my polygot file causing the tablebases not to load. The games will be re-run. And maybe we'll get a chance to see how much the tables help it.
User avatar
abik
Posts: 819
Joined: Fri Dec 01, 2006 10:46 pm
Location: Mountain View, CA, USA
Full name: Aart Bik

Re: The great 40/120 Tournament

Post by abik »

krazyken wrote:It appears there was an error in my polygot file causing the tablebases not to load. The games will be re-run. And maybe we'll get a chance to see how much the tables help it.
Thanks for letting me know Kenny! I am glad BikJump's lookup code is not at fault here. Now I "just" have to fix the king safety issue somehow (I always find this the most tricky part of the evaluation, balancing computational time and accuracy).