Vajolet2 2.4 Release

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

fastgm
Posts: 818
Joined: Mon Aug 19, 2013 6:57 pm

Re: Vajolet2 2.4 Release

Post by fastgm »

Over the weekend I will do a test with 13 threads and enable the LittleBlitzer logging (Log First Engine).
If this test runs without time losses I will repeat it with 15 threads.
elcabesa
Posts: 855
Joined: Sun May 23, 2010 1:32 pm

Re: Vajolet2 2.4 Release

Post by elcabesa »

thank you,
if it's possible please take not of cpu utilization in a tournament where Vajolet is playing and one without Vajolet.
Dann Corbit
Posts: 12540
Joined: Wed Mar 08, 2006 8:57 pm
Location: Redmond, WA USA

Re: Vajolet2 2.4 Release

Post by Dann Corbit »

NPS by core count from 1 to 11:

994817
1962355
2649980
3803760
4601974
5311562
5689381
5814181
6392830
6609065
7016777

I will run some high core counts when I get home tonight, unless I forget.
;-)
Taking ideas is not a vice, it is a virtue. We have another word for this. It is called learning.
But sharing ideas is an even greater virtue. We have another word for this. It is called teaching.
fastgm
Posts: 818
Joined: Mon Aug 19, 2013 6:57 pm

Re: Vajolet2 2.4 Release

Post by fastgm »

I've done a few tests. Unfortunately, I was unable to create a debug log, because the LittleBlitzer GUI crashed after a few games, if I enabled the feature "Log First Engine".
The cpu utilization shows no difference between running a tournament with or without Vajolet.

1. Test
LittleBlitzer GUI, TC=60"+0.6", 13 parallel games, 61 opponents
976 games, 26 time losses

2. Test
LittleBlitzer GUI, TC=60"+0.6", 15 parallel games, 2 opponents
700 games, 22 time losses

3. Test
LittleBlitzer GUI, TC=60"+0.6", 8 parallel games, 2 opponents
210 games, 9 time losses

4. Test
LittleBlitzer GUI, TC=100"+1", 15 parallel games, 61 opponents
1532 games, 7 time losses

5. Test
Cutechess-Cli, TC=60"+0.6", 15 parallel games, 2 opponents
1000 games, no time losses

So it seems it's up to the LittleBlitzer GUI.
Perhaps the "move overhead" settings in Vajolet are a bit too low/aggressive, compared to the other engines?

Best wishes,
Andreas
elcabesa
Posts: 855
Joined: Sun May 23, 2010 1:32 pm

Re: Vajolet2 2.4 Release

Post by elcabesa »

usually I don't use any move overhead with cutechess. you can try with some small move overhead
fastgm
Posts: 818
Joined: Mon Aug 19, 2013 6:57 pm

Re: Vajolet2 2.4 Release

Post by fastgm »

Move overhead isn't neccesary with Cutechess, see 5. Test.
With Cutechess Vajolet runs fine without time losses.

5. Test
Cutechess-Cli, TC=60"+0.6", 15 parallel games, 2 opponents
1000 games, no time losses

The LittleBlitzer GUI has no option "move overhead" and as stated before, other engines do not have time losses, execpt Vajolet2 2.4
Is it possible to implement an UCI option in your engine? Stockfish, Komodo and Houdini for example has this option.
elcabesa
Posts: 855
Joined: Sun May 23, 2010 1:32 pm

Re: Vajolet2 2.4 Release

Post by elcabesa »

i didn't noticed that 5. was executed using cutechess.

are you using some kind of move overhead for other eingines?
I'll do some testings
fastgm
Posts: 818
Joined: Mon Aug 19, 2013 6:57 pm

Re: Vajolet2 2.4 Release

Post by fastgm »

There are only a few engines that have an option like "move overhead".
For Houdini 6, I had to set this option to 30.
option name Move Overhead type spin min 0 max 5000 default 0

All other engines runs fine with the default settings.