Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / CFish 9 (BMI2 compile)
- - By Eelco de Groot (***) Date 2018-01-31 20:53 Edited 2018-01-31 20:56
Attachment: cfish.exe (201k)
Attachment: src.7z - Directly from Ronald's Git, no changes except compressing by 7z. (110k)
Parent - - By leavenfish (***) [us] Date 2018-02-16 02:27
So, what distinguishes Cfish from...Stockfish or even ASMfish?
Parent - - By user923005 (****) [us] Date 2018-02-16 06:48
CFish is faster than SF and fairly well synchronized.
AsmFish is also faster but for unknown reasons it is also weaker than SF now.
http://www.sp-cc.de/

P.S.
Brainfish is just stockfish with a special book.
Parent - - By leavenfish (***) [us] Date 2018-02-17 06:05
What are those known reasons?

I understands various compiles are always fast than Stockfish....but still don't really know what sets them apart. Simply speed?
Parent - By user923005 (****) [us] Date 2018-02-20 19:29
The ports are intended as direct translations.  Nobody knows what happened to AsmFish to reduce its strength.
More speed normally means more Elo.
Parent - - By Labyrinth (*****) [us] Date 2018-02-16 22:29
Stockfish is written in C++

CFish is Stockfish written in C

ASMFish is Stockfish written in assembly language.

Theoretically, using C can result in a faster, more optimized program than C++ at the cost of some coding difficulty/complexity.

Theoretically, using assembly as opposed to C or C++ can result in an even faster, even more optimized program at the cost of a tremendous increase in coding difficulty/complexity.
Parent - - By chess666 (*) [fr] Date 2018-02-17 08:00
Hello,

With a simple Intel i7 Skylake, i cannot see a difference with CFish ( even, the Brainfish compile is faster ! ) except for asmFish but it is not updated.....

By the way, McBrain seem the same as Stockfish 9 except some options ? I don't know why they say it is for LTC as correspondance players ???

Many thanks
Parent - - By jpqy (**) [be] Date 2018-02-17 08:14
asmFish is already updated with a few patches these last days!

https://github.com/lantonov/asmFish/tree/executables
Parent - - By chess666 (*) [fr] Date 2018-02-17 08:40 Edited 2018-02-17 08:54
Yes but it is the 2017-12-03 Stockfish version, not the last one !
Parent - - By jpqy (**) [be] Date 2018-02-17 09:37
Everyone who follow asmFish knows that..it was more then 2month ago that asmFish got any update..so they are busy for people who don't know!
Parent - - By chess666 (*) [fr] Date 2018-02-17 09:58 Edited 2018-02-17 10:03
It was not a criticism, just a fact, no worry ;-)
My goal is to have informations about engines based on Stockfish, particularly McBrain because i understand what the others do, no more !
Parent - - By leavenfish (***) [us] Date 2018-02-17 17:03
Of course the only important thing is: is that older version of Stockfish actually stronger or weaker than the current version. Improvements seem mighty hard to come by  (save for the 'contempt' trick for engine play)
Parent - By chess666 (*) [fr] Date 2018-02-17 18:28
Yes, it is not easy to see the progress but anyway, SF9 is stronger than SF8 and so, the progress exist .....even with contempt=0 ;-)
So, nobody know something about McBrain please ?
Parent - By Labyrinth (*****) [us] Date 2018-02-18 02:29
Hadn't heard of McBrain until you mentioned it. The wiki states the opposite of what you described, stating that it "includes a number of Stockfish development patches that passed short time control (STC) but failed LTC", meaning it is better suited for short time controls. As to how true that is I haven't the foggiest.
Parent - - By McBrain [us] Date 2018-02-18 05:19
Hi , I just joined this forum and will be happy to answer questions you might have about McBrain.  McBrain has evolved, and the current version, besides being updated for SF9, was enhanced to solve many positions that SF will never get , or will take a very long to get.  Most of the ideas I am using in the current version failed at STC and the time controls that fishtest uses 10/sec game with 0.1 sec  increment for STC and 1 minute game and .6 sec increment would make them almost impossible to pass either STC to LTC on fishtest.  Also, ,many of the patches have been modified as well and are not the exact same patch that was submitted to fishtest (although some are).  If the patch helped solve a certain type of position, that is what I was looking for, while trying my best to keep it close to SF at the longer tc's.

As an example , try these four positions in SF 9 and compare the result to McBrain 9. 

7k/pp2Np1p/2p2P1N/5pP1/7K/P4n2/1q3Q2/5n2 w - - 0 1


8/6pp/1K6/N5P1/3N4/8/npn1P3/k7 w - - 0 1


4r1rk/p3qpp1/1pnp1n1p/5P2/P1PPP3/4Q2P/2BB2R1/6RK w - - 0 1


8/8/8/3p2p1/kppN1pP1/2P2P2/1PR2K2/2q5 b - - 0 1


I describe McBrain as an alternative engine, one that can be used in conjunction with Komodo, Houdini or SF 9 , not as a replacement engine.  I would definitely consider using McBrain in very LTC games .  Generally speaking , anything time control longer than 2 min with 4 sec increment on relatively high end machine ( 2M nps or higher) McBrain will hold its own against SF 9, the faster tc's, SF will dominate.  It is almost impossible to say what will happen at very long tc's - but I am very confident McBrain would do just fine.

https://github.com/MichaelB7/Stockfish/releases/tag/v9
Parent - - By leavenfish (***) [us] Date 2018-02-18 06:50
"I describe McBrain as an alternative engine, one that can be used in conjunction with Komodo, Houdini or SF 9 , not as a replacement engine.  I would definitely consider using McBrain in very LTC games .  Generally speaking , anything time control longer than 2 min with 4 sec increment on relatively high end machine ( 2M nps or higher) McBrain will hold its own against SF 9, the faster tc's, SF will dominate.  It is almost impossible to say what will happen at very long tc's - but I am very confident McBrain would do just fine."

In other words...not appropriate (compared to the other engines you mention) to actual LTC (2 min + 4 sec...laughable to really be thought of as LTC)...or analysis?
Parent - - By chess666 (*) [fr] Date 2018-02-18 08:02 Edited 2018-02-18 08:46
I agree leanvenfish BUT it is just what i search because i am a correspondence player !
However, Stockfish has yet many issues and it is why we use too Houdini and Komodo ( and some things with our own brain ;-) )

@McBrain,
OK, i have your engine just for complete , you are right and i hope you will resolve a new problem of Stockfish which is "dynamic contempt" ???
Indeed, with normal contempt ( at 20 in SF9 ), i put 0 ( or off in McBrain ) and that's all BUT the dynamic one is hard coded and i cannot remove it !!!
Stocfish team don't see the interest --> https://groups.google.com/forum/#!topic/fishcooking/RAak9c9h0VE
just because they think about engines games ! BUT for me, even TCEC is not a long time control, and so i need the exact evaluation because i don't play against weaker engines and don't need artificial tricks ( i can see myself when a move choice will be draw in many cases and don't need contempt for know if i accept the draw or take the risk to win)  ....

Hope you will can solve this issue ?

Kind regards
Jacques
Parent - By leavenfish (***) [us] Date 2018-02-18 17:29
I was 2399 ICCF about 15 yrs ago when I noticed it was becoming harder for me to win...of course I was playing in a Wch Semi Final so the competition was tough. Anyway, I quit rather than go up against engines. But I wonder sometime what it would be like today if I used engines. It's just an artificial kind of chess anymore (correspondence)...don 't think I would enjoy it with everyone using engines these days.

So, I just use engines to shore up different opening repertoires I use for different OTB play - long normal time controls...or the G/45 which is so prevalent locally. So, I only care about...going into lines where my opponent has MORE chances to go wrong...a tighter rope to walk so to speak. Engines showing multi-pv are good for that so it really does not matter how 'fast' an engine is to me, but I like to keep up with the different ones, but really ones is as good as the other if you use multi-pv, it really is. Again, for my purposes.
Parent - - By McBrain [us] Date 2018-02-18 21:33
There is no dynamic contempt in McBrain and I seriously doubt if McBrain will ever have that code.  That code was introduced after Stockfish 9 was released.  What McBrain does have as default setting , when in analysis mode , contempt = zero.  That is changeable via the uci interface.  The setting is "Analysis Contempt"  and when it is off, the contempt setting equals zero in analysis mode regardless of the contempt numeric setting.  You can also set so that contempt  white , black or both in analysis or both- meaning it switches sides with each move.  Not sure why somebody would want to use the "both" setting, but it's there if one wants to use. it.
Parent - - By chess666 (*) [fr] Date 2018-02-18 22:12 Edited 2018-02-18 22:38
I know that ! I was speaking for the future update where the dynamic contempt hard coded made after SF9 wiil be present if they don't remove it ???

Finally, your have two BMI2 compile ! What's the difference please ?

Many thanks
Parent - - By McBrain [us] Date 2018-02-18 23:06
I have several friends that make exe's for me ,  from my own experience , not one machine  and not one complier make the fastest exe for all machines.  You just reminded me that I forgot to place one set of exe's out there.   I have to run right now, but when I get back will place them out on GitHub and post here.  Functionally , they are the same, just use the one that gives you the highest nps number on your machine when running the "bench" command.

Since i have a fork, I get to pick and choose and what goes in and what comes out.  I'm not crazy about all of the simplification patches they do, but I understand it's part of the process.  I tend to delay them for a while.  Also patches that take away problem solving are a pet peeve.  My biggest challenge are the modifications that I have.  You would think I could just copy them forward, but that is not always the case
Parent - - By leavenfish (***) [us] Date 2018-02-19 00:10
Like most if not all these engine 'programmers'...they are groping in the dark...adding this, removing that, test...rise and repeat and hope to get a little something in the end. IMHO...
Parent - By chess666 (*) [fr] Date 2018-02-19 11:39
Yes, i agree !
The positions above just show us a bug in Stockfifh 9 BUT if you use it with multi-PV, no problem to find them !
As i use always multi PV ( and even some tricks ), i don't see the difference between McBrain and SF9 ???
Parent - - By McBrain [us] Date 2018-02-18 21:19
No it's meant for very long time controls or analysis, I just don't care if it does poorly at what I call bullet or micro time controls.  Your mileage may vary naturally.
Parent - By leavenfish (***) [us] Date 2018-02-19 00:11
And I for one than you for that!
Up Topic The Rybka Lounge / Computer Chess / CFish 9 (BMI2 compile)

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill