Winboard Forum

There are some news on the fruitchess website:

http://www.fruitchess.com

regards
Joachim

Future Plans:

Development on Fruit has ceded. This page stays for archival purposes.

Team / History:

Ryan Benitez joined the Fruit-Team in December 2005. Before he developed Gambit-Fruit with some interesting changes on the basis of the last open-source-version 2.1. He contiues to work on Fruit.

Which of these is correct?

I don't know. Both statements are speculations on my side.

To be more specific: I think Fabien has ceded development on Fruit, whereas I think that Ryan still works on it. I'll clarify that on the page.

Joachim Rang wrote:There are some news on the fruitchess website:...

Firstly thank you very much for these great engines.

Secondly (impertinent): is there any chance of a Linux compile of the version from 05/11/03?

Regards

Volker

I forgot who did the Linux compile for version 2.2.1. I think it was Dann Corbit and I could ask him, or do you know anyone trustable who can make such a compile?

Joachim Rang wrote:I forgot who did the Linux compile for version 2.2.1. I think it was Dann Corbit and I could ask him...

Good idea IMO.

Joachim Rang wrote:... or do you know anyone trustable who can make such a compile?

Natürlich alle mit Stirnglatze, Krawatte und Jackett aus Bern !

Seriously: if you trust me I can try it myself. Of course I 'll not abuse the source or give it to anyone else.

Volker

Volker Pittlik wrote:Seriously: if you trust me I can try it myself. Of course I 'll not abuse the source or give it to anyone else.

Most likely this won't be very difficult. Compiling Fruit 2.1 was only a matter of saying "make".

Note however that the Windows version under wine runs a little faster than my own Linux compile. This is in spite of the fact that the Windows version is 32 bit and my Linux version is 64 bit.

For Fruit 2.2 the difference between the Windows and the Linux version is even larger. In the initial position it's 400 kn/s for the Linux version and 500 kn/s for the Windows version with wine. I think both are 32 bit.

Onno Garms wrote:

Volker Pittlik wrote:Seriously: if you trust me I can try it myself. Of course I 'll not abuse the source or give it to anyone else.

Most likely this won't be very difficult. Compiling Fruit 2.1 was only a matter of saying "make".

Note however that the Windows version under wine runs a little faster than my own Linux compile. This is in spite of the fact that the Windows version is 32 bit and my Linux version is 64 bit.

For Fruit 2.2 the difference between the Windows and the Linux version is even larger. In the initial position it's 400 kn/s for the Linux version and 500 kn/s for the Windows version with wine. I think both are 32 bit.

The 4.x versions of GCC have profile guided optimization. If you do that, I guess it will be about 25-30% faster.

Also, if you choose flags for your specific system you can also get a boost from that.

See:
http://www.dis.com/gnu/gcc/Optimize-Options.html

Onno Garms wrote:...
Most likely this won't be very difficult. Compiling Fruit 2.1 was only a matter of saying "make"...

Yes. The only difficulties I noticed have been in the 1.3.4 version of the Toga variation which tries to use Daniel's bitbases. But after changing some libraries and an additional linker option it at least compiles without problems.

Onno Garms wrote:...Note however that the Windows version under wine runs a little faster than my own Linux compile. This is in spite of the fact that the Windows version is 32 bit and my Linux version is 64 bit...

What prevents me to use engines with wine at the moment (besides aesthetic considerations ) is that I noticed that all of them freeze or hang after a while. I usually test at fast time controls with hundreds of games per engine. I found no specific reason for it. It happens to all windows engines I tested therefore I assume its a problem of wine. I can only say that it happens preferential shortly after I stoped observing so that as much games as possible are not played. Doesn't that happen to you?

Volker

Volker Pittlik wrote:What prevents me to use engines with wine at the moment (besides aesthetic considerations ) is that I noticed that all of them freeze or hang after a while.

Have you tried Rybka 64-bit with microwine? Version 0.2 apparently fixes several crashes, but it hasn't been stress-tested.

Teemu Pudas wrote:...
Have you tried Rybka 64-bit...

I neither have the engine nor a 64-bit OS .

Teemu Pudas wrote:... with ... but it hasn't been stress-tested.

I have great sympathies for tested programs. Also I prefer (besides chess) packages coming out one the the "official" repositories for the Linux distributions. Microwine is not.

The problem is not so serious. Sometimes it works for hundred of games, another time it crashes every game. Unpredictable.

Volker

Volker Pittlik wrote:I usually test at fast time controls with hundreds of games per engine. I found no specific reason for it. It happens to all windows engines I tested therefore I assume its a problem of wine. I can only say that it happens preferential shortly after I stoped observing so that as much games as possible are not played. Doesn't that happen to you?

I also run hundreds of games with fast time control. I use Shredder Windows GUI with wine and both Linux and Windows engines. In approx. 10.000 games I had exactly one hanging. This occured when I tried to run tests and go on a weekend trip. Normally I stop and restart tests after at most one day because I need my computer for something else. When I return after a day, I always find the tests still running.

I run wine-0.9.2. With later versions of wine, the Shredder GUI crashes from time to time.

Dann Corbit wrote:The 4.x versions of GCC have profile guided optimization. If you do that, I guess it will be about 25-30% faster.

Thank you for pointing out. I think this hint will be a big help for my future compiles.

However, my observersion on Fruit 2.2 refers to the precompiled versions. I think it was you who compiled them.

Dann Corbit wrote:...The 4.x versions of GCC have profile guided optimization. If you do that, I guess it will be about 25-30% faster.

Also, if you choose flags for your specific system you can also get a boost from that...

Joachim have been so kind to send me the Fruit source code. I got some speed-up from the profile guided optimization in gcc but only 10-15%. Best results I got from compilations with the Intel compiler (with Interprocedural Optimization (IPO) and Profile-guided Optimization (PGO) enabled). It takes some experiments to find the best combination. Change in nps was from ~520 knps to ~765 knps on one cpu on my E6300.

I made a test with both version against some opponents.

compiled with gcc

Code: Select all

Rank Name                   Elo    +    - games score oppo. draws
   1 Fruit 2.2.1             68   29   28   312   63%   -20   34%
   2 Fruit (Toga) 1.2.1a     67   48   48   100   51%    60   42%
   3 Fruit 05/11/03          51   30   29   298   60%   -21   30%
   4 Glaurung2 e4 perf        7   48   49   100   42%    60   39%
   5 Spike 1.2 Turin        -14   50   51   102   39%    59   27%
   6 Ruffian 2.1.0          -51   52   54   102   35%    59   21%
   7 Scorpio 1.91           -60   50   52   102   32%    59   31%
   8 Shredder Classic 1.3   -67   50   52   104   31%    59   30%


compiled with the Intel compiler

Code: Select all

Rank Name                   Elo    +    - games score oppo. draws
   1 Fruit 05/11/03         105   34   33   240   69%   -26   32%
   2 Fruit (Toga) 1.2.1a     83   52   52    80   51%    77   50%
   3 Fruit 2.2.1             49   33   32   240   61%   -26   36%
   4 Glaurung2 e4 perf      -20   53   55    80   34%    77   48%
   5 Spike 1.2 Turin        -20   56   58    80   36%    77   29%
   6 Shredder Classic 1.3   -23   58   61    80   36%    77   20%
   7 Scorpio 1.91           -38   56   58    80   33%    77   33%
   8 Ruffian 2.1.0         -137   61   67    80   21%    77   24%


I send the executable to Joachim who will possible put it on the Fruit website.

Volker

P.S. About the source code: Please ask Fabien or Joachim.

I wonder why Toga is not listed with its proper name: It is not the original Fruit 2.1 engine, nor a clone (because it is clearly stronger than the original from which it is derived). Instead, its authors have managed to keep it the strongest open source engine for quite a while now in spite of being open source. So I think it deserves to be called by its own name.

Fritz Grau wrote:I wonder why Toga is not listed with its proper name... its authors have managed to keep it the strongest open source engine for quite a while now in spite of being open source. So I think it deserves to be called by its own name.

I think the Fruit/Toga issue has been discussed so often that everyone had a fair chance to find his point of view. I accept yours, here comes mine:

I did my best to compile the fastest executables of Toga and the different Fruit versions. The Linux version of Fruit is available from the official server. The Toga version compiled with the Intel compiler got also a significant speed-up. I didn't publish that version and I have no plans to do so. Everyone can download the Intel compiler for free and make the fastest thing for his machine. That is what I recommend.

If I understand the GPL correctly I'm allowed to modify the program as I like as long I don't distribute it. While I made several binaries of it to get the fastest executable I changed some lines so that they reflect the circumstances as I see it. When started from the command line Toga's output looks as follow here:

Code: Select all

volker@vpittlik:~/schach/toga> ./fruit-toga
Fruit (Toga) 1.2.1a by Fabien Letouzey. Modifications by Thomas Gaksch.
uci
id name Fruit (Toga) 1.2.1a
id author Fabien Letouzey. Modifications by Thomas Gaksch


BTW: The makefile coming with the Toga source code always named the executable as "fruit". I always renamed it to "fruit-toga". Maybe someone will complain even about that.

From my point of view -YMMV- the one who deserves the greatest respect is Fabien Letouzey and no one else.

Now I have found out it the test that the Fruit version of 05/11/03 is also the strongest fruit variant here at my computer. Others came to similar results. (There have been a discussion about it at CSS-Forum some days ago.) That is the version I will use for my tests in the future.

Volker

Thank you for your reply. I was not aware that you just used the official name chosen by Toga itself because I have always run it under an interface (Arena) which hides the raw UCI output by default.

> From my point of view -YMMV- the one who deserves the greatest respect is Fabien Letouzey and no one else.

It was not my intention to challenge your views or to decide who deserves most respect. I believe hardly anybody disagrees about the view that Fruit is a milestone, while Toga is something like a "tuned version" of this engine. I never had the patience to read a license agreement, but I always thought that the license of Fruit made Toga legal.

I do dislike what I experience as mobbing against Toga, but as you explained the origin of the name in your list (of which I was ignorant), it is obvious that your choice of name is no part of it.

Regards, F.