The Week in Chess

Tuesday, July 30, 2013

Bouquet 1.7b x64 vs. Ivanhoe 1945a x64 rematch

I have conducted a rematch between Bouquet 1.7b x64 and Ivanhoe 1945a x64 due to the comments made by Velmarin. In the Bouquet 1.7b run, where Bouquet 1.7b was eventually ranked very high at #8 in the rating list, I made a disclaimer that it failed to win a single match in my AMD Quad core which is the primary chess engines tournament computer with a minimum of 4 simultaneous matches running.

The computers used where Bouquet 1.7b scored favorably were all AMD dual cores. But there were some rematches made because of irregular scores. I traced the dual core computer that has the irregular scores and found out that it was the powerful of the 3 dual cores with 4 simultaneous tournaments running at a time while the other 2 has 3 tournaments.

In the Ivanhoe 1945a run, Bouquet 1.7b lost to Ivanhoe with a wide margin. I did not bother to look at the details of the result in the haste to publish the Ivanhoe run right after it was finished and the many favors I have done for Bouquet 1.7b. When the comments of Velmarin and Dmitri Gusev came out regarding the losses by time forfeit, I checked on the details of the tournament and found that the time forfeits were more than the 5% of the personal limit I set.  I traced the AMD dual core computer that caused, it was again the powerful one with 4 tournaments running.

After some reflection, I noticed a similarity of the problem I encountered with Stockfish which was published "Stockfish 3 Killed While Sleeping on Threads". The findings was that Stockfish get worse with more tournaments running simultaneously and the resolution was enabling the "Use Sleeping Threads" option in the UCI, which was also pointed out by Marco Costalba. This was independently verified and confirmed by Dmitri that the threads problem exist.

It is Velmarin's turn to check on the problem by running it in a computer with many processess/threads running to verify his Bouquet engine's reliability in computers with heavy processing load. Hopefully, Velmarin can reproduce the problem I identified.

Upon reconsideration, I decided to do a rematch between Bouquet 1.7b and Ivanhoe 1945a to clear some garbage.

These are the results of the rematch, the first score will be used and will appear in the next rating list issue.

Bouquet 1.7b x64 vs. Ivanhoe 1945a – Rematch
CPU: AMD Dual Core – 5000 bogomips, 4GB
OS: Linux/Windows 7
GUI: Arena 3.0
Time control: 1 minute + 1 second

Losses by

Time Forfeit
Description Score% Win Loss Draw Total Bouquet Ivanhoe
Bouquet-Ivanhoe Re-match 1 62.00 33 9 58 100 0 1
Bouquet-Ivanhoe Re-match 2 61.50 31 8 61 100 0 2
Bouquet-Ivanhoe Re-match 3 62.50 36 11 53 100 1 2
Total 62.00 100 28 172 300 1 5

Download the results of the above tournament games here.

1 comment:

  1. Dear John Wilson,

    Thank you very much for investigating the time forfeit phenomenon in more detail!



Chessdom News