Nirvanachess 2.0 Release

cetormenter · Post by **cetormenter** » Wed Dec 17, 2014 10:04 am

A complete rewrite of Nirvana has been done. This brought the engine into the modern age by converting to bitboards. This also allowed me to clean up a lot of the code and will make future improvements even easier. The biggest part of this release is addition of multithreading support. This was done via the lazy smp method described elsewhere on this board.

Some comparisons to other engines/version 1.8

Gauntlet against various engines at 6" + 0.05"

Code: Select all

Rank Name              Elo    +    - games score oppo. draws 
   1 Senpai            326   37   37   392   82%    56   14% 
   2 Nirvanachess2.0   134   25   25   656   61%    47   18% 
   3 SmarThink1.6       46   30   30   390   49%    56   20% 
   4 Nirvanachess1.8     0   21   21   911   45%    48   18% 
   5 Arasan17.3        -51   31   31   393   36%    56   18% 
   6 Fruit            -130   33   33   392   26%    56   20%

Direct match up at 60" + 0.05"

Code: Select all

Rank Name              Elo    +    - games score oppo. draws 
   1 Nirvanachess2.0   198   39   39   275   77%     0   21% 
   2 Nirvanachess1.8     0   39   39   275   23%   198   21%

Gauntlet at 60" + 0.05"

Code: Select all


Rank Name              Elo    +    - games score oppo. draws 
   1 Senpai             57   55   55   114   58%     0   25% 
   2 Nirvanachess2.0     0   31   31   455   69%  -156   19% 
   3 Arasan17.3       -175   59   59   114   26%     0   21% 
   4 SmarThink        -184   61   61   113   26%     0   17% 
   5 Fruit            -323   74   74   114   13%     0   11%

Frank Quisinsky · Post by **Frank Quisinsky** » Wed Dec 17, 2014 10:11 am

Hi Thomas,

fantastic.
I will switched in my still running tourney to NirvanaChess 2.0.
Do you saw that NirvanaChess 1.8 have a really big Problem in KingSafty compare to version 1.7. Your version 1.8 lost a lot of games very very fast. That is the reason that version 1.7 is around 10-20 Elo stronger as version 1.8.

Thanks for your new version.
I will delete the 300 games I have for version 1.8 in my FCT1 - 2015 - tourney and will replay the games with version 2.0.

Best
Frank

cetormenter · Post by **cetormenter** » Wed Dec 17, 2014 10:23 am

Yes this is due to the lazy evaluation, which gained the engine a lot of speed at the expense of having a less accurate king safety function. This wasn't an issue in self play but as you saw, and as I later saw, in gauntlet play this is actually a fairly large regression. This has been removed in this version.

As always, thank you very much for your testing. Its very good to see how Nirvana preforms under various conditions and against various opponents.

Frank Quisinsky · Post by **Frank Quisinsky** » Wed Dec 17, 2014 10:27 am

Hi Thomas,

nice if you have fun with my work ...

After the still running Equinox 3.30 x64 round Robin (Need around 6.5 days, started in the night) your version 2.0 will replay the games from version 1.8.

An other Problem I have with Tornado 6 in my still running FCT1 - 2015 - tourney. Version 5.0 is clearly stronger.

Tornado development is stopped. I will replaced Tornado 6.0 with DisasterArea and in this case two changes I have to make if I restart this tourney in 6.5 days.

Very interesting in testing NirvanaChess ... I am waiting of your new version, available now after the big work you do in the last months!

Thanks again!

Best
Frank

Graham Banks · Post by **Graham Banks** » Wed Dec 17, 2014 10:40 am

cetormenter wrote:A complete rewrite of Nirvana has been done. This brought the engine into the modern age by converting to bitboards. This also allowed me to clean up a lot of the code and will make future improvements even easier. The biggest part of this release is addition of multithreading support. This was done via the lazy smp method described elsewhere on this board.

Some comparisons to other engines/version 1.8

Gauntlet against various engines at 6" + 0.05"
Code: Select all
Rank Name              Elo    +    - games score oppo. draws 
   1 Senpai            326   37   37   392   82%    56   14% 
   2 Nirvanachess2.0   134   25   25   656   61%    47   18% 
   3 SmarThink1.6       46   30   30   390   49%    56   20% 
   4 Nirvanachess1.8     0   21   21   911   45%    48   18% 
   5 Arasan17.3        -51   31   31   393   36%    56   18% 
   6 Fruit            -130   33   33   392   26%    56   20% 
Direct match up at 60" + 0.05"
Code: Select all
Rank Name              Elo    +    - games score oppo. draws 
   1 Nirvanachess2.0   198   39   39   275   77%     0   21% 
   2 Nirvanachess1.8     0   39   39   275   23%   198   21% 
Gauntlet at 60" + 0.05"
Code: Select all
Rank Name              Elo    +    - games score oppo. draws 
   1 Senpai             57   55   55   114   58%     0   25% 
   2 Nirvanachess2.0     0   31   31   455   69%  -156   19% 
   3 Arasan17.3       -175   59   59   114   26%     0   21% 
   4 SmarThink        -184   61   61   113   26%     0   17% 
   5 Fruit            -323   74   74   114   13%     0   11% 

Thanks Thomas.

Dr.Wael Deeb · Post by **Dr.Wael Deeb** » Wed Dec 17, 2014 4:54 pm

Excellent Thomas

So were talking about an improvement of 100-120 Elo comparing to version 1.8

Dr.D

Wolfgang · Post by **Wolfgang** » Wed Dec 17, 2014 7:00 pm

problem here: does NOT run on AMD-FX6300 (six core)

no problems on Intel (older C2D, i5-520, i7)

could you provide a compile which runs on AMD too?
would be nice...

same problem as @Gabor: readme says "three" different versions, but rar-file only contains one .exe which is not limited to modern systems but to Intel (seemingly)...

Could anyone with similar (AMD) hardware countercheck? Ray?

cdani · Post by **cdani** » Wed Dec 17, 2014 8:31 pm

Works on AMD-FX 8350. Did not try a tournament for the moment.

cetormenter · Post by **cetormenter** » Wed Dec 17, 2014 8:45 pm

Dr.Wael Deeb wrote:Excellent Thomas

So were talking about an improvement of 100-120 Elo comparing to version 1.8
Dr.D

Hopefully! We shall see soon on the rating lists. Just in raw speed there quite a significant gain. Not to mention the bug fixes to the evaluation function.

Gabor Szots wrote: Thomas, I have found only one exe in the package. Is that 64-bit? It does run on my Q6600, which makes me wonder.

Yes there is only a 64 bit compile but I suppose it doesn't hurt to create a 32 bit compile too.

Wolfgang Battig wrote:
PostPosted: Wed Dec 17, 2014 6:00 pm Post subject: Re: Nirvanachess 2.0 Release
problem here: does NOT run on AMD-FX6300 (six core) Sad
no problems on Intel (older C2D, i5-520, i7)

could you provide a compile which runs on AMD too?
would be nice... Wink

same problem as @Gabor: readme says "three" different versions, but rar-file only contains one .exe which is not limited to modern systems but to Intel (seemingly)...

Could anyone with similar (AMD) hardware countercheck? Ray?

I think the AMD cpu must not have access to the bitscan functions. I have reuploaded this time including both a 64 bit compile without those functions and a 32 bit compile. This does incur a slight speed drop however it is only about 2%. Hopefully this fixes the issue. If not then I am at a loss.

Link:
http://www.mediafire.com/download/i7ohp ... ss_2.0.rar

Wolfgang · Post by **Wolfgang** » Wed Dec 17, 2014 10:21 pm

Thanks Thomas!

It seems to be a problem with my AMD-FX only. On another AMD (X2-270) it works, no problem...

The new compile also does not work on the FX, so it really seems to be an issue (configuration?!) with this machine