Crafty 22.0 tests

[beachknight]

Hi Tony,

First thing I did with a new crafty release is:

running it in console mode and typing perf in command-line.

I see the figures for a moment and repeat it
few times if necesssary. This was without rc file.
I repeat this perf typing with rc. Then quit. repeat
this process for bench. Since total nodes figure
is not constant, I follow raw nodes per second
figure for comparison purposes.

For all JA builds, bench caused a premature
termination of console mode of crafty.

Best,

[beachknight]

Crafty 22.0 JA w32

Code: Select all

   Engine         Score     Cr    Cr    S-B
1: Crafty 22.0 JA 2.5/5  ····· 1===0    6,25
1: Crafty 22.0 PS 2.5/5  0===1 ·····    6,25

5 of 10 games played
Name of the tournament: CET17SSiSS66a07
Site/ Country: SSiSS, Antalya, Türkiye
Level: Blitz 4/4

Crafty 22.0 JA :
book learning +
log files: +


seems ok with corrected rc.

[bob]

Arena 1.99b4 GUI, tablebases off for 22.0

First game: 2+2
Remaining: 3+3
Last two: with mt=2 as clp (does not help further)

crafty.rc

Code: Select all

mt=2
ponder off
#egtb
#tbpath=d:\tb
#cache=32M
hash=48M
hashp=16M
#swindle on
learn 7
book random 1
bookw freq 0.7
bookw ratio 0.8
bookw eval 0.6
bookw learn 1
bookw cap 0.5
book width 4
#book on
#show book
#smpmt=2
#timebook 80 8
mode=tournament
computer
#log=on
exit

Code: Select all

CET17SSiSS66a02  2008

                                      1234567890
1   Crafty 21.06 PS Blended P3  3667   ½½½½011101   6.0/10
2   Crafty 22.0 JA2             3700  ½½½½100010    4.0/10

two more games: 22.0 forfeits on time.

Someone asked me to look at this thread because of the above note. And that indicates something is up, I regularly play 20-30 thousand games at a time, and over the past 2,000,000 games I have played, crafty has not lost a single game on time. It was not clear what time control you use, but if you have any sort of increment, then crafty losing on time indicates a setup problem because it normally will not do that. A log.nnn file where it loses on time would be useful to see what is going on...

Further blitz tests with crafty.rc

Code: Select all

mt=2
ponder off
egtb
tbpath=d:\tb
cache=32M
hash=48M
hashp=16M
#swindle on
learn 7
book random 1
#bookw freq 0.7
#bookw ratio 0.8
#bookw eval 0.6
#bookw learn 1
#bookw cap 0.5
#book width 4
#book on
#show book
#smpmt=2
#timebook 80 8
mode=tournament
computer
#log=on
exit

[bob]

Hi Tony,

First thing I did with a new crafty release is:

running it in console mode and typing perf in command-line.

I see the figures for a moment and repeat it
few times if necesssary. This was without rc file.
I repeat this perf typing with rc. Then quit. repeat
this process for bench. Since total nodes figure
is not constant, I follow raw nodes per second
figure for comparison purposes.

For all JA builds, bench caused a premature
termination of console mode of crafty.

Best,

what do you mean "total nodes figure is not constant"???

If that is true, you have a bug unless you are doing SMP tests, then node counts can vary a lot...

[beachknight]

Hi,

Simplified rc file below solved forfeit problems.

Code: Select all

hash=48M
hashp=10M
mt=2
egtb
tbpath=d:\tb
cache=16M
ponder off
swindle on
exit

I think this problem occured because
Crafty 22.0 JA was pondering, while playing
against another pondering Crafty, ie 21.06,
that does not free 1 CPU after moving, ie
50% of total CPU. mode tournament and kibitz 1
was then included in rcs.

[bob]

Hi,

Simplified rc file below solved forfeit problems.

Code: Select all

hash=48M
hashp=10M
mt=2
egtb
tbpath=d:\tb
cache=16M
ponder off
swindle on
exit

OK. more processes than processors will wreck my SMP search and can easily make it step over a fairly fast time control. But for longer games, I can't imagine how it would lose on time...

I think this problem occured because
Crafty 22.0 JA was pondering, while playing
against another pondering Crafty, ie 21.06,
that does not free 1 CPU after moving, ie
50% of total CPU. mode tournament and kibitz 1
was then included in rcs.

[beachknight]

Code: Select all

   Engine         Score          Cr
1: Spike 1.1      6.5/10 =11====011 
2: Crafty 22.0 PS 3.5/10 ·········· 

10 games played / Tournament finished
Name of the tournament: CET17SSiSS66d
Site/ Country: SSiSS, Antalya, Türkiye
Level: Blitz 2/2



Crafty with pondering, no book learning

Spike 1.1 without pondering, no book learning 


Crafty.rc

hash=48M
hashp=10M
mt=2
egtb
tbpath=d:\tb
cache=16M
swindle on
book on
log on
ponder on
exit

[beachknight]

Code: Select all

   Engine          Score       Cr
1: Crafty 22.0 PS  3.5/7  ······· 
1: Crafty 22.0 JA2 3.5/7  ====01= 

7 of 10 games played
Name of the tournament: CET17SSiSS66e
Site/ Country: SSiSS, Antalya, Türkiye
Level: Blitz 2/2

Crafty 22.0 PS  with pondering, with book learning 

Crafty 22.0 JA2 with pondering, no book learning 

I'll change the book of Crafty 22.0 JA2 for remaining
three games.

[beachknight]

Code: Select all

   Engine          Score          Cr
1: Crafty 22.0 JA2 5.5/10 ====01==1= 
2: Crafty 22.0 PS  4.5/10 ·········· 

10 games played / Tournament finished
Name of the tournament: CET17SSiSS66e
Site/ Country: SSiSS, Antalya, Türkiye
Level: Blitz 2/2

Just replaced the PS book with JA one.

Crafty 22.0 JA2 : book learning +

With pondering without problem so far.