Critter 0.80

[Don]

Hi Richard,

How to get a 64 bit linux binary?

Don

Hi,

unfortunately I am stuck with the Microsoft compiler for now, my code does not compile with GCC due to problems with C++ templates. So there will be no linux binaries in the near future.

Richard

Edit: Last time I checked, win32 binaries worked fine with WINE, but I don't know about 64bit.

Hi Richard,

Thanks for the response.

As someone has noted, there is an Intel compiler for Linux, I'm actually about to check it out for myself soon.

Don

[LucenaTheLucid]

Hi again Richard, some more results from your newest Critter:

Code: Select all

    Program                          Elo    +   -   Games   Score   Av.Op.  Draws

  9 Naum 4.2                       : 2948   43  43   160    48.8 %   2957   36.2 %
 10 Critter 0.80                   : 2947   45  45   150    48.0 %   2961   34.7 %
 11 Critter 0.70                   : 2894   53  54   117    32.9 %   3017   33.3 %
 12 Komodo 1.2                     : 2881   44  45   160    38.8 %   2961   33.8 %

It seems to have leap frogged Komodo and caught up to Naum. And always of course the error bars are still very high. Very promising results though.

You can follow my tournament here:

http://www.open-chess.org/viewtopic.php ... 1&start=20

[rvida]

Richard,

Do you have the possibility to compile a POPCNT binary for us with Core i7?

Best Regards,
Martin

Well, I am able to make a POPCNT capable binary, but I'm afraid it will be not very useful (only 1-2% nps increase). Recently I did some profiling, and the most prominent bottleneck in my engine seems to be in the hash table accesss. I believe that a well placed "prefetch" instruction will show a significantly higher nps gain. Either way, I would like to implement these optimizations in the next release.

Richard

[Ozymandias]

What's the gtb64.dll file for?
Scorpio bitbases?

[Trahald]

Gaviota Tablebases

[Matthias Gemuh]

Gaviota Tablebases

64 Bit

[Frank Quisinsky]

Hello Herbert,

not all problems are so easy to find. In this case its easy because I have it in different games.

I am sure, in SWCR databases programmers can found a lot of such problems, because I playing up to mate without resign. OK, database game average = 85 moves (without resign in ATL-4 or IPON ... Ingo wrote about it) = 67 moves. So my games need around 20% more time.

Hello Richard,

good, you found it.
In 1 1/2 weeks the next SWCR-64 round I will start. It would be great (my wish) if you can create a small update, perhaps 0.80a.

Thanks again for Critter.

Best
Frank

[Martin Thoresen]

Well, I am able to make a POPCNT capable binary, but I'm afraid it will be not very useful (only 1-2% nps increase). Recently I did some profiling, and the most prominent bottleneck in my engine seems to be in the hash table accesss. I believe that a well placed "prefetch" instruction will show a significantly higher nps gain. Either way, I would like to implement these optimizations in the next release.

Richard

You mean you will release a POPCNT binary in the next release? If so, that's great news. Even though it's only 1-2-3% it is definitely an increase.

[Martin Thoresen]

Also Richard, you should either update your current homepage or make a new one.

[swami]

Neatly "Optimized and fine tuned" engine. Critter has always shown better progress in almost nearly every single STS epd's ever since version 0.3 (?).

Congrats, Richard. Delighted to find that it's playing even better than Zappa, Fritz or Naum!

Once STS Stat is updated, I'll test this in STS 11 and STS 12. As of now, STS Stat 2.5 doesn't process the results from last 2 EPD sets.

Critter 0.80



In comparison with Critter 0.70



http://sites.google.com/site/strategict ... st-results

1000 Positions
10 seconds per position
Hardware: Q6600, 32 bits, 2 GB RAM, 2.4 GHZ. Arena 2.01 GUI.


Best Wishes,
Swami