Page 1 of 4

Komodo and SF: small android tourney

Posted: Tue Dec 30, 2014 9:09 pm
by Laskos
I pitted the two strongest android engines SF 121014 and Komodo 8 on my Samsung Galaxy Tab 4, SF gets about 700kn/s on its 4 cores.

2 engines on 4 cores, 2 on 1 core, round-robin 60 games total:

Code: Select all

                                            POINTS      ELO

SF 121014  4 threads    *   6.5  7.0  7.0   20.5/30    +133
Komodo 8   4 threads   3.5   *   6.0  7.5   17.0/30     +47
SF 121014  1 thread    3.0  4.0   *   7.0   14.0/30     -23
Komodo 8   1 thread    3.0  2.5  3.0   *     8.5/30    -161
SF 121014 seems the strongest, and the improvement from 1 to 4 cores is similar to that one can see on an i5 or i7 system at short TC.

Re: Komodo and SF: small android tourney

Posted: Wed Dec 31, 2014 9:46 am
by Laskos
After a total of 120 games in round-robin:

Code: Select all

                                               POINTS      ELO 

SF 121014  4 threads    *   13.5  14.5  16.5   44.5/60    +184 
Komodo 8   4 threads   6.5    *   10.5  15.0   32.0/60     +23 
SF 121014  1 thread    5.5   9.5   *    13.5   28.5/60     -17 
Komodo 8   1 thread    3.5   5.0   6.5   *     15.0/60    -191

Re: Komodo and SF: small android tourney

Posted: Wed Dec 31, 2014 10:10 am
by Adam Hair
Thanks, Kai. I had thought about doing a test like this, but it could not have been with 4 cores.

So, SF continues to show its superiority on smaller hardware and/or short time controls, while it appears Komodo surpasses it on top end hardware at long time controls.

Re: Komodo and SF: small android tourney

Posted: Wed Dec 31, 2014 11:00 am
by Laskos
Adam Hair wrote:Thanks, Kai. I had thought about doing a test like this, but it could not have been with 4 cores.

So, SF continues to show its superiority on smaller hardware and/or short time controls, while it appears Komodo surpasses it on top end hardware at long time controls.
Komodo seems to suffer more on 32 bits (my Tab is 32 bit 4.4.2 KitKat OS) than SF. Komodo reaches ~500kn/s to SF's ~700kn/s on 4 cores, while on 64 bits the speeds are similar. TC is 1s/move, equivalent to 0.1s/move on i5 or i7, so very short and not so efficient TC. I will wait for Aart's Bik "Chess for Android" new TC management promised by Aart to be released very soon.

Re: Komodo and SF: small android tourney

Posted: Sat Jan 03, 2015 1:17 pm
by Laskos
Six best multi-core Android engines on 4 cores, 1 sec/move in "Chess for Android", round-robin with a total 300 games:

Code: Select all

               POINTS      ELO

SF 121014      85.0/100    +301
Komodo 8       65.5/100    +111
Critter 1.6a   58.0/100     +56
BlackMamba 2.0 49.0/100      -7
IvanHoe 9.46b  22.5/100    -215
Gaviota v1.0   20.0/100    -241
And the final between SF and Komodo, 100 games:

Code: Select all

               POINTS      ELO

SF 121014      71.0/100     +78
Komodo 8       29.0/100     -78

Re: Komodo and SF: small android tourney

Posted: Sat Jan 03, 2015 11:02 pm
by stavros
hi. iam curious if the 12/10/2014 droidfish is the best for android cos i saw this:http://chesstroid.blogspot.gr/2014/12/s ... again.html
seems droidfish 1.54(sf 5) is better than latest android compile (1.55,12/10/2014) !!

Re: Komodo and SF: small android tourney

Posted: Sun Jan 04, 2015 8:14 am
by Gurcan Uckardes
Yes. SF5 works better. Some say 64 games are statistical noise and object that but they forget one thing: SF5 is compiled under Linux using MingW. I guess SF dev version is not and compiling environment matters a lot!
stavros wrote:hi. iam curious if the 12/10/2014 droidfish is the best for android cos i saw this:http://chesstroid.blogspot.gr/2014/12/s ... again.html
seems droidfish 1.54(sf 5) is better than latest android compile (1.55,12/10/2014) !!

Re: Komodo and SF: small android tourney

Posted: Sun Jan 04, 2015 8:19 am
by Gurcan Uckardes
Kai, check game terminations of Black Mamba. It has a bad habit of 'engine exits' while it's winning. I have many games i had to correct from 0-1 to 1-0 or vice versa.
Laskos wrote:Six best multi-core Android engines on 4 cores, 1 sec/move in "Chess for Android", round-robin with a total 300 games:

Code: Select all

               POINTS      ELO

SF 121014      85.0/100    +301
Komodo 8       65.5/100    +111
Critter 1.6a   58.0/100     +56
BlackMamba 2.0 49.0/100      -7
IvanHoe 9.46b  22.5/100    -215
Gaviota v1.0   20.0/100    -241
And the final between SF and Komodo, 100 games:

Code: Select all

               POINTS      ELO

SF 121014      71.0/100     +78
Komodo 8       29.0/100     -78

Re: Komodo and SF: small android tourney

Posted: Sun Jan 04, 2015 12:35 pm
by Laskos
Gurcan Uckardes wrote:Yes. SF5 works better. Some say 64 games are statistical noise and object that but they forget one thing: SF5 is compiled under Linux using MingW. I guess SF dev version is not and compiling environment matters a lot!
stavros wrote:hi. iam curious if the 12/10/2014 droidfish is the best for android cos i saw this:http://chesstroid.blogspot.gr/2014/12/s ... again.html
seems droidfish 1.54(sf 5) is better than latest android compile (1.55,12/10/2014) !!
Strange, I just tested now in 200 games on 4 cores, but at short TC of 1 sec/move:

Code: Select all

                       POINTS      ELO

SF 121014-armv7      106.0/200     +10
SF 5-armv7            94.0/200     -10
Maybe it's your longer time control, maybe just too few games.

Re: Komodo and SF: small android tourney

Posted: Sun Jan 04, 2015 12:39 pm
by Laskos
Gurcan Uckardes wrote:Kai, check game terminations of Black Mamba. It has a bad habit of 'engine exits' while it's winning. I have many games i had to correct from 0-1 to 1-0 or vice versa.
Laskos wrote:Six best multi-core Android engines on 4 cores, 1 sec/move in "Chess for Android", round-robin with a total 300 games:

Code: Select all

               POINTS      ELO

SF 121014      85.0/100    +301
Komodo 8       65.5/100    +111
Critter 1.6a   58.0/100     +56
BlackMamba 2.0 49.0/100      -7
IvanHoe 9.46b  22.5/100    -215
Gaviota v1.0   20.0/100    -241
And the final between SF and Komodo, 100 games:

Code: Select all

               POINTS      ELO

SF 121014      71.0/100     +78
Komodo 8       29.0/100     -78
Gurcan, it's you with that monster test with tens of thousands games at 5 sec/move? Would make for a good rating list. Maybe a bit tough on your Note II :)

Just inspected the PGN file of my tourney of 6 best engines, BlackMamba exited just once, so the results won't change much. Thanks for bringing out to me about the problem.