Bestia 0.3H verses Quark 1.15 Alpha

[Dann Corbit]

Geschrieben von: / Posted by: Dann Corbit at 05 February 2001 20:51:28:
Quark 1.15alpha gets 2/3 of the points from Bestia on a long time control match (G/60 or better) on fast machines. Here is the outcome of several machines running over the weekend:
Program Elo + - Games Score Av.Op. Draws
--------- ---- -- -- ----- ------- ------- -------
1 Quark : 2361 53 70 110 66.8 % 2239 20.9 %
2 Bestia : 2239 70 53 110 33.2 % 2361 20.9 %
Games are available from here:
ftp://cap.connx.com/pub/Bestia-Quark.pgn
I felt compelled to run a long experiment, since I made some assertions on Friday which were really based primarily on "gut-feel" and not on very much data. So (for my setup at least) my suspicions are confirmed.
This version of Quark is not publicly available, but I believe that T.M. has plans to introduce a formal release at the early part of February.
The Quark program is making incredible progress. The only think it can be likened to is Yace.
I should also comment on the remarkable progress of Bestia. There has obviously been an incredible amount of work done on this program as well.
The version of Quark I am running is probably the strongest, because I don't think the others (with greater version numbers) were created with the Intel compiler. For Quark, there is a 50% performance penalty or more on high end machines for using other compilers.
Quark is a 4-crown engine and {prediction} will get points at Paderborn.


My FTP Site

[Christian Söderström]

Geschrieben von: / Posted by: Christian Söderström at 05 February 2001 21:23:58:
Als Antwort auf: / As an answer to: Bestia 0.3H verses Quark 1.15 Alpha geschrieben von: / posted by: Dann Corbit at 05 February 2001 20:51:28:

Quark 1.15alpha gets 2/3 of the points from Bestia on a long time control match (G/60 or better) on fast machines. Here is the outcome of several machines running over the weekend:
Quark is a 4-crown engine and {prediction} will get points at Paderborn.

Indeed! I don't know exactly what a 4-crown engine is but indeed Quark will probably steal a couple of points in Paderbord!
I remember the good old days when Thomas was still sorting out the basics and Mint was able to beat Quark. Yesterday I ran a test match between the new Mint v2.0 and Quark 1.13 and the result was a rather comfortable 8-0 to Quark I'm not a strong chess player but to me Quarks play looked impressive. Mint could usually hold its own in the opening and often in the middlegame, but when it came to endgames and especially transitions to endgames Quark made a clean sweep. I know that Thomas has always told me he put a lot of work in the pawn evaluation in endgames, and now I believe him
Of course, it could just be that Mint is really bad in endgames, but I don't see how that would make anybody happy, so I prefer my former theory.
Cheers!
- Christian Söderström

Mint Homepage

[Dann Corbit]

Geschrieben von: / Posted by: Dann Corbit at 05 February 2001 21:51:36:
Als Antwort auf: / As an answer to: Re: Bestia 0.3H verses Quark 1.15 Alpha geschrieben von: / posted by: Christian Söderström at 05 February 2001 21:23:58:

Quark 1.15alpha gets 2/3 of the points from Bestia on a long time control match (G/60 or better) on fast machines. Here is the outcome of several machines running over the weekend:
Quark is a 4-crown engine and {prediction} will get points at Paderborn.

Indeed! I don't know exactly what a 4-crown engine is but indeed Quark will probably steal a couple of points in Paderbord!
I remember the good old days when Thomas was still sorting out the basics and Mint was able to beat Quark. Yesterday I ran a test match between the new Mint v2.0 and Quark 1.13 and the result was a rather comfortable 8-0 to Quark I'm not a strong chess player but to me Quarks play looked impressive. Mint could usually hold its own in the opening and often in the middlegame, but when it came to endgames and especially transitions to endgames Quark made a clean sweep. I know that Thomas has always told me he put a lot of work in the pawn evaluation in endgames, and now I believe him
Of course, it could just be that Mint is really bad in endgames, but I don't see how that would make anybody happy, so I prefer my former theory.
Cheers!

I was playing some mint games the other day, and I noticed that instead of the usual !,?, !!, ?! notations, mint uses different smilies like , and I thought that was really funny. Maybe mint needs a winkie when it does something funny/tricky.


My FTP Site

[Christian Söderström]

Geschrieben von: / Posted by: Christian Söderström at 05 February 2001 22:15:13:
Als Antwort auf: / As an answer to: Re: Bestia 0.3H verses Quark 1.15 Alpha geschrieben von: / posted by: Dann Corbit at 05 February 2001 21:51:36:

Quark 1.15alpha gets 2/3 of the points from Bestia on a long time control match (G/60 or better) on fast machines. Here is the outcome of several machines running over the weekend:
Quark is a 4-crown engine and {prediction} will get points at Paderborn.

Indeed! I don't know exactly what a 4-crown engine is but indeed Quark will probably steal a couple of points in Paderbord!
I remember the good old days when Thomas was still sorting out the basics and Mint was able to beat Quark. Yesterday I ran a test match between the new Mint v2.0 and Quark 1.13 and the result was a rather comfortable 8-0 to Quark I'm not a strong chess player but to me Quarks play looked impressive. Mint could usually hold its own in the opening and often in the middlegame, but when it came to endgames and especially transitions to endgames Quark made a clean sweep. I know that Thomas has always told me he put a lot of work in the pawn evaluation in endgames, and now I believe him
Of course, it could just be that Mint is really bad in endgames, but I don't see how that would make anybody happy, so I prefer my former theory.
Cheers!

I was playing some mint games the other day, and I noticed that instead of the usual !,?, !!, ?! notations, mint uses different smilies like , and I thought that was really funny. Maybe mint needs a winkie when it does something funny/tricky.

Yes, those are the legendary smilievals Unfortunately a unique feature. Mint is the only program I know that exploits the very flexible PV output of winboard, except for various (Book: 56%) stuff of course. For the upcoming release of Mint (codenamed "Ginger" - which just drew Bestia 0.3h!) I plan to implement Smilieval 2.0 (tm).
The faces have been around since the early days of Millennia (my old chess program). Incidentally, a Winboard version of Millennia is probably one of the rarest creatures in the Winboard-universe
And while I'm thinking of the good old days, has everybody seen the legendary match between Millennia and Beroz Dehdari?
http://ds.onyx.nu/man-vs-machine/
Goodnight everyone!
- Christian Söderström

Mint Homepage

[WYx]

Geschrieben von: / Posted by: WYx at 06 February 2001 10:18:33:
Als Antwort auf: / As an answer to: Bestia 0.3H verses Quark 1.15 Alpha geschrieben von: / posted by: Dann Corbit at 05 February 2001 20:51:28:

Quark 1.15alpha gets 2/3 of the points from Bestia on a long time control match (G/60 or better) on fast machines. Here is the outcome of several machines running over the weekend:
Program Elo + - Games Score Av.Op. Draws
--------- ---- -- -- ----- ------- ------- -------
1 Quark : 2361 53 70 110 66.8 % 2239 20.9 %
2 Bestia : 2239 70 53 110 33.2 % 2361 20.9 %
Games are available from here:
ftp://cap.connx.com/pub/Bestia-Quark.pgn
I felt compelled to run a long experiment, since I made some assertions on Friday which were really based primarily on "gut-feel" and not on very much data. So (for my setup at least) my suspicions are confirmed.
This version of Quark is not publicly available, but I believe that T.M. has plans to introduce a formal release at the early part of February.
The Quark program is making incredible progress. The only think it can be likened to is Yace.
I should also comment on the remarkable progress of Bestia. There has obviously been an incredible amount of work done on this program as well.
The version of Quark I am running is probably the strongest, because I don't think the others (with greater version numbers) were created with the Intel compiler. For Quark, there is a 50% performance penalty or more on high end machines for using other compilers.
Quark is a 4-crown engine and {prediction} will get points at Paderborn.

[U.Tuerke]

Geschrieben von: / Posted by: U.Tuerke at 06 February 2001 11:06:20:
Als Antwort auf: / As an answer to: Bestia 0.3H verses Quark 1.15 Alpha geschrieben von: / posted by: Dann Corbit at 05 February 2001 20:51:28:

Quark 1.15alpha gets 2/3 of the points from Bestia on a long time control match (G/60 or better) on fast machines. Here is the outcome of several machines running over the weekend:
Program Elo + - Games Score Av.Op. Draws
--------- ---- -- -- ----- ------- ------- -------
1 Quark : 2361 53 70 110 66.8 % 2239 20.9 %
2 Bestia : 2239 70 53 110 33.2 % 2361 20.9 %
Games are available from here:
ftp://cap.connx.com/pub/Bestia-Quark.pgn
I felt compelled to run a long experiment, since I made some assertions on Friday which were really based primarily on "gut-feel" and not on very much data. So (for my setup at least) my suspicions are confirmed.
This version of Quark is not publicly available, but I believe that T.M. has plans to introduce a formal release at the early part of February.
The Quark program is making incredible progress. The only think it can be likened to is Yace.
I should also comment on the remarkable progress of Bestia. There has obviously been an incredible amount of work done on this program as well.
The version of Quark I am running is probably the strongest, because I don't think the others (with greater version numbers) were created with the Intel compiler. For Quark, there is a 50% performance penalty or more on high end machines for using other compilers.
Quark is a 4-crown engine and {prediction} will get points at Paderborn.

Do I get you right Dann, you say, that the engine compiled with the Intel compiler is 50 percent faster than say the msvc compiled version ?
Thx, Uli

[Thomas Mayer]

Geschrieben von: / Posted by: Thomas Mayer at 06 February 2001 12:02:48:
Als Antwort auf: / As an answer to: Re: Bestia 0.3H verses Quark 1.15 Alpha geschrieben von: / posted by: U.Tuerke at 06 February 2001 11:06:20:
Hi Uli,

The version of Quark I am running is probably the strongest, because I don't
think the others (with greater version numbers) were created with the Intel
compiler. For Quark, there is a 50% performance penalty or more on high end
machines for using other compilers.

Do I get you right Dann, you say, that the engine compiled with the Intel
compiler is 50 percent faster than say the msvc compiled version ?

My own compiler is only a MSVC Standard and optimization-features are very low there... - So Danns Code is in nearly double that fast then the code I can produce...
Greets, Thomas
P.S.: And so Dann will compile my paderborn version...

[U.Tuerke]

Geschrieben von: / Posted by: U.Tuerke at 06 February 2001 13:46:14:
Als Antwort auf: / As an answer to: Re: Bestia 0.3H verses Quark 1.15 Alpha geschrieben von: / posted by: Thomas Mayer at 06 February 2001 12:02:48:

Hi Uli,

The version of Quark I am running is probably the strongest, because I don't
think the others (with greater version numbers) were created with the Intel
compiler. For Quark, there is a 50% performance penalty or more on high end
machines for using other compilers.

Do I get you right Dann, you say, that the engine compiled with the Intel
compiler is 50 percent faster than say the msvc compiled version ?

My own compiler is only a MSVC Standard and optimization-features are very low there... - So Danns Code is in nearly double that fast then the code I can produce...
Greets, Thomas
P.S.: And so Dann will compile my paderborn version...

Where to download the compiler ? -:)
Uli

[Thomas Mayer]

Geschrieben von: / Posted by: Thomas Mayer at 06 February 2001 14:40:58:
Als Antwort auf: / As an answer to: Re: Bestia 0.3H verses Quark 1.15 Alpha geschrieben von: / posted by: U.Tuerke at 06 February 2001 13:46:14:
Hi Uli,

Where to download the compiler ? -:)

you can download a test-version direct on the intel-site. But I think it only will work if you have the MSVC++ professional... And if you have the professional-version your gain will not be as high as mine, maybe 5-10 % but I think it depends on the processor... - And I think Dann do some more complicate work to optimize all at its best - and sure, some routines from Quark are with Danns compilition double the speed then my version... incredible ! You can imagine who I was looking on my screen the first time Dann sends me a Corbit-version of my program... it takes some minutes until I could close my mouth...
Greets, Thomas

[Dann Corbit]

Geschrieben von: / Posted by: Dann Corbit at 06 February 2001 20:27:05:
Als Antwort auf: / As an answer to: Re: Bestia 0.3H verses Quark 1.15 Alpha geschrieben von: / posted by: U.Tuerke at 06 February 2001 11:06:20:
[snip]

Do I get you right Dann, you say, that the engine compiled with the Intel compiler is 50 percent faster than say the msvc compiled version ?

There is a large variation, and it depends very much on the engine in question. For instance, every chess engine I compile performs better than even the MS VC++ professional C++ compiler version 6.0 with all patches and the processor pack, except for one: Arasan. I have no idea why I cannot speed Arasan up. For some engines it is dramatic. If you are using GCC or the Borland compiler, I guarantee an enormous speedup. Same for any MS compiler below the professional (We have 'Enterprise' here).
With the Intel compiler, there are a lot of things you can tweek also. Using profile guided optimization, I can sometimes get another dramatic speedup.

My FTP Site