Sting 4.7, one of Stockfish derivatives, is ported to Android, thanks to Jim Ablett. New compiles do not seem to stop raining and this can only make us happier.
The result of 3 minutes test with Exynos 4412 running at 1.6 Ghz x 4 cores is 579 kNps which is way lower than Stockfish. But of course, that doesn't directly point to any weakness. Remember the rule: Nodes perfo only speaks between different hardware running the same compile of a given engine.
My plan about Sting is to play only one test against Stockfish 5 and ignore it in Rapidroid ranking for the moment. I'm not very interested in derivatives yet.
The binary can be downloaded from my engine collection: HERE
HAL9000
"It just isn't conceivable that you can design a program strong enough to beat players like me."
January 8, 2015
Android UCI engine update: Senpai 1.0.JA
In addition to the version released by Fabien Letouzey, the developer, we have now an alternative compile of Senpai 1.0 made by Jim Ablett.
Jim compiles by default with arm v5TE instruction set. The initial compile is v5 too but different in size. I won't have time to conduct a head-to-head comparison of both compiles due to new engines that need gauntlets. Any elo gap between two compiles should be related to different compiler efficiencies. Apparently, JA compile gives more nodes per second performance, as shown below:
The JA compile is available: HERE in my repository and also at Jim Ablett's site
Jim compiles by default with arm v5TE instruction set. The initial compile is v5 too but different in size. I won't have time to conduct a head-to-head comparison of both compiles due to new engines that need gauntlets. Any elo gap between two compiles should be related to different compiler efficiencies. Apparently, JA compile gives more nodes per second performance, as shown below:
JA compile reaches 352 kNps, about %70 more than the original!
Official compile reaches 204 kNps
The JA compile is available: HERE in my repository and also at Jim Ablett's site
Android XB engine update: Crafy 24.1.JA without logging
No more logs... at last
There have been a lot of complaints about the log files automatically created by Crafy compile. The usual config file of Crafty didn't work under Android. Consequently, the engine folder used to get quickly populated by several log files, two per each game played.
Thanks to Jim Ablett, we have now a new compile of latest Crafty which won't generate logs anymore. Yummy, isn't it?
You may download the Crafty 24.1.JA no logs binary: HERE
January 6, 2015
Strongest Android engine: Stockfish to investigate once again
Yes i've been daring to state Stockfish 5 is still the strongest of all, even better than Stockfish 121014, the more recent version, on Android OS, in contrast to Windows. Surprise, surprise!
The conclusion was based on 192 games played by two Stockfishes and Komodo 8, using 32 superfinal openings of TCEC-6, both sides.
The device was an Asus 176CX running a quad core Intel Z3745D. And of course, all 3 engines were compiled for i686 instructions.
Komodo 8 could not stand against the fishes and SF5 had beaten both the dev version as well as Komodo.
Although i'm confident about the result, some friends didn't like it too much and stated it should be "statistical noise", a technical term used as an emergency argument :-)
Finally we live in a digital century where even 64 games are not enough to convince.
I wasn't interested at all in playing 10000 games to confirm again SF5 plays better on Android. But a bullet test conducted by Kai Laskos on Exynos cpu showed 121014 version better.
The only remarkable difference of his test is the time control of 1 sec vs 15 sec of my test.
The cpu architecture can also (slightly?) affect the result but i'd checked knps performances and they looked stable and proportional.
Thus, i reopen the case now to discover again which one is better, to convince everyone.
Let's go crazier this time: It will be 908 games at 15 sec/move, on Exynos 4412 cpu. All 454 openings used in TCEC-6, played with white and black.
The conclusion should be definitive this time and the games will start next week. I expect 908 games need 22 full days.
Patience please, stay tuned and have mercy for my poor Exynos!
The conclusion was based on 192 games played by two Stockfishes and Komodo 8, using 32 superfinal openings of TCEC-6, both sides.
The device was an Asus 176CX running a quad core Intel Z3745D. And of course, all 3 engines were compiled for i686 instructions.
Komodo 8 could not stand against the fishes and SF5 had beaten both the dev version as well as Komodo.
Although i'm confident about the result, some friends didn't like it too much and stated it should be "statistical noise", a technical term used as an emergency argument :-)
Finally we live in a digital century where even 64 games are not enough to convince.
I wasn't interested at all in playing 10000 games to confirm again SF5 plays better on Android. But a bullet test conducted by Kai Laskos on Exynos cpu showed 121014 version better.
The only remarkable difference of his test is the time control of 1 sec vs 15 sec of my test.
The cpu architecture can also (slightly?) affect the result but i'd checked knps performances and they looked stable and proportional.
Thus, i reopen the case now to discover again which one is better, to convince everyone.
Let's go crazier this time: It will be 908 games at 15 sec/move, on Exynos 4412 cpu. All 454 openings used in TCEC-6, played with white and black.
The conclusion should be definitive this time and the games will start next week. I expect 908 games need 22 full days.
Patience please, stay tuned and have mercy for my poor Exynos!
Android UCI engine update: Discocheck 5.2.1
Well, since Jim Ablett restarted to compile for Android, engines started to fall like rain. And to be honest, this kind of rain doesn't need an umbrella at all.
The last update coming is Discocheck 5.2.1 over the last stable version 4.3.
Lucas Braesch stopped working on Discocheck long time ago in order to become one of the maintainers of Stockfish. This, we may call 5.2.1 the final version.
You may download the Android binary: HERE
The last update coming is Discocheck 5.2.1 over the last stable version 4.3.
Lucas Braesch stopped working on Discocheck long time ago in order to become one of the maintainers of Stockfish. This, we may call 5.2.1 the final version.
You may download the Android binary: HERE
January 4, 2015
Android engines collection: 4 new versions added
The updated versions of 4 engines, compiled by Jim Ablett can be downloaded from my collection too. Enjoy!
On devices with Intel cpu's however, they simply don't work. As explained before, even though some arm compiles can work with Intel, the instruction translations cost node speeds as low as 1/3. Sorry for Inteldroids...
To download more Android engines, the entire collection is available HERE.
Arasan 17.4.JA: click HERE to download
Crafty 24.1.JA: click HERE to download
Rodent 1.6.JA: click HERE to download
I've tested them all on Galaxy Note II. They are compiled for arm v5TE instruction set. No issues when running under Chess for Android.
On devices with Intel cpu's however, they simply don't work. As explained before, even though some arm compiles can work with Intel, the instruction translations cost node speeds as low as 1/3. Sorry for Inteldroids...
To download more Android engines, the entire collection is available HERE.
January 3, 2015
Jim Ablett is back
That famous scene... totally recalled
For those who did not know him, he's a legendary contributor into chess engine development thanks to his compiles not only for Android but also Windows too.
Glad to see you're back Jim.
First pack of recent versions are available at:
http://jimablett.net63.net/compiles.html
* Crafty 24.1 JA
* Arasan 17.4 JA
* Scorpio 2.7.7 JA
* Rodent 1.6 JA (added 4-Jan-15)
Now, i can't wait for others to come :-)
Subscribe to:
Posts (Atom)












