Firenzina 2.2.2 xTreme vs. Stockfish 2.2.2 (sse42-ja-intel)

[Gusev]

I posted the first long-control (40/40) 20-game match between Firenzina 2.2.2 xTreme (the original /MD compile) and Stockfish 2.2.2 (sse42-ja-intel) at http://dgusev.cs.edinboro.edu/GameDev/C ... _40-40.zip. Both engines were restricted to 6 threads each. The test suite was nunn.pgn. The match was drawn, +4-4=12. Naturally, no reliable conclusion about Firenzina's strength can be drawn from such a short match. It is posted not to make any claims, but merely to encourage people to test Firenzina some more.

The /MT compile not requiring dlls from Microsoft Visual C++ 2012 Redistributable Package (x64) is now available at http://dgusev.cs.edinboro.edu/GameDev/C ... MT_x64.zip.

As I mentioned before, when I compiled Stockfish 2.2.2 and Gull II using the settings of Firenzina's original /MD compile, the resulting engines were noticeably weaker than their factory compiles. This indicates to me that there's room for improvement here.

For the time being, Firenzina is hosted at http://dgusev.cs.edinboro.edu/GameDev/C ... nzina.html. Will move at some point soon.

[geots]

I posted the first long-control (40/40) 20-game match between Firenzina 2.2.2 xTreme (the original /MD compile) and Stockfish 2.2.2 (sse42-ja-intel) at http://dgusev.cs.edinboro.edu/GameDev/C ... _40-40.zip. Both engines were restricted to 6 threads each. The test suite was nunn.pgn. The match was drawn, +4-4=12. Naturally, no reliable conclusion about Firenzina's strength can be drawn from such a short match. It is posted not to make any claims, but merely to encourage people to test Firenzina some more.

The /MT compile not requiring dlls from Microsoft Visual C++ 2012 Redistributable Package (x64) is now available at http://dgusev.cs.edinboro.edu/GameDev/C ... MT_x64.zip.

As I mentioned before, when I compiled Stockfish 2.2.2 and Gull II using the settings of Firenzina's original /MD compile, the resulting engines were noticeably weaker than their factory compiles. This indicates to me that there's room for improvement here.

For the time being, Firenzina is hosted at http://dgusev.cs.edinboro.edu/GameDev/C ... nzina.html. Will move at some point soon.

Dmitri- a hypothetical here. I just downloaded both the MT compile and the other also. I have no idea yet which will work or won't work. But I have a couple intel i5 4-core systems- NO HP- running in Windows 7. Let's just suppose for argument's sake THEY WILL BOTH WORK. In that case, which of the 2 would you use if you were in my place?



Best,

george

[kranium]

Dmitri- a hypothetical here. I just downloaded both the MT compile and the other also. I have no idea yet which will work or won't work. But I have a couple intel i5 4-core systems- NO HP- running in Windows 7. Let's just suppose for argument's sake THEY WILL BOTH WORK. In that case, which of the 2 would you use if you were in my place?
Best,
george

Hi George-
i hope you don't mind me answering for Dimitri, but
i suggest: for both compiles...

double-click on executable
type 'benchmark' at console prompt
record total NPS

use whichever is faster on your system

the /MD compile might need
Microsoft Visual C++ 2012 Redistributable Package (x64)
(vcredist_x64.exe)
http://www.microsoft.com/en-us/download ... x?id=30679
if you don't already have it on your system

Norm

[geots]

Dmitri- a hypothetical here. I just downloaded both the MT compile and the other also. I have no idea yet which will work or won't work. But I have a couple intel i5 4-core systems- NO HP- running in Windows 7. Let's just suppose for argument's sake THEY WILL BOTH WORK. In that case, which of the 2 would you use if you were in my place?
Best,
george

Hi George-
i hope you don't mind me answering for Dimitri, but
i suggest: for both compiles...

double-click on executable
type 'benchmark' at console prompt
record total NPS

use whichever is faster on your system

the /MD compile might need
Microsoft Visual C++ 2012 Redistributable Package (x64)
(vcredist_x64.exe)
http://www.microsoft.com/en-us/download ... x?id=30679
if you don't already have it on your system

Norm

Thank you Norm- in fact I am glad you answered. You hear it from the expert- the answer is not debatable. Let me take this opportunity- as I think you are right on the edge considering it- and all it might take is a light push- to ask you to seriously consider getting back in the game with 2.3. It is just this engine, meaning 2.2 xTreme, can play at times like a world champion. And I don't care what Dmitri or anyone else does- they can't match what you can do. And that is not a criticism of him- or anyone else. He would agree with me as well.



Thanks again- (and consider it please),

george

[geots]

Dmitri- a hypothetical here. I just downloaded both the MT compile and the other also. I have no idea yet which will work or won't work. But I have a couple intel i5 4-core systems- NO HP- running in Windows 7. Let's just suppose for argument's sake THEY WILL BOTH WORK. In that case, which of the 2 would you use if you were in my place?
Best,
george

Hi George-
i hope you don't mind me answering for Dimitri, but
i suggest: for both compiles...

double-click on executable
type 'benchmark' at console prompt
record total NPS

use whichever is faster on your system

the /MD compile might need
Microsoft Visual C++ 2012 Redistributable Package (x64)
(vcredist_x64.exe)
http://www.microsoft.com/en-us/download ... x?id=30679
if you don't already have it on your system

Norm

The decision was made for me. The MD version will not work because of the "missing dll" that is not missing. This particular dll, I have been at war with for quite a while. This has come up a number of times. Last was with one of velmarin's Bouquet versions. The dll I don't have is actually in my system in 3 different places. Out of frustration, I once copied and pasted a couple of them into the engine folder right next to the executable and said "There, you blind son of a bitch!" Then I tried to run the exe in console mode again, and again it said I needed the dll in my system.



Grrrrr-

george

[Gusev]

The decision was made for me. The MD version will not work because of the "missing dll" that is not missing. This particular dll, I have been at war with for quite a while. This has come up a number of times. Last was with one of velmarin's Bouquet versions. The dll I don't have is actually in my system in 3 different places. Out of frustration, I once copied and pasted a couple of them into the engine folder right next to the executable and said "There, you blind son of a bitch!" Then I tried to run the exe in console mode again, and again it said I needed the dll in my system.

Grrrrr-

george

George,
"I feel your pain" (c) Bill Clinton. Does the MT compile work for you? If so, you're in good shape. These two compiles just ran to a shockingly exact draw in my 500-game 1/1 test,
http://www.talkchess.com/forum/viewtopi ... 01&t=47388
I am now testing a variation of MT with multiple additional settings recommended to me by the pros. For future references, I may call this modified set of parameters SGVAD (Schmidt-Gusev-Velasco-Ablett-Dart) after all involved: In this order of letters, just so it reminds "squad". kNodes/sec visibly improved, and it remains to be seen to which extent ELO tracks. I attribute most of the improvement to the PGO (Profile-Guided Optimization) strongly recommended by Jim Ablett.

[Gusev]

I think you are right on the edge considering it- and all it might take is a light push- to ask you to seriously consider getting back in the game with 2.3. It is just this engine, meaning 2.2 xTreme, can play at times like a world champion. And I don't care what Dmitri or anyone else does- they can't match what you can do. And that is not a criticism of him- or anyone else. He would agree with me as well.

For the record, yes, I totally agree with George! The road is open for you, Norm, to either release Fire 2.3, or join Firenzina GitHub and work on Firenzina 2.3, also with you as the primary author, as it should be. Assuming that you get along with Jose Velasco who gave the advice on how one could make Fire stronger (http://www.talkchess.com/forum/viewtopi ... 00&t=47152), I'd like to invite him to the team.

[geots]

The decision was made for me. The MD version will not work because of the "missing dll" that is not missing. This particular dll, I have been at war with for quite a while. This has come up a number of times. Last was with one of velmarin's Bouquet versions. The dll I don't have is actually in my system in 3 different places. Out of frustration, I once copied and pasted a couple of them into the engine folder right next to the executable and said "There, you blind son of a bitch!" Then I tried to run the exe in console mode again, and again it said I needed the dll in my system.

Grrrrr-

george

George,
"I feel your pain" (c) Bill Clinton. Does the MT compile work for you? If so, you're in good shape. These two compiles just ran to a shockingly exact draw in my 500-game 1/1 test,
http://www.talkchess.com/forum/viewtopi ... 01&t=47388
I am now testing a variation of MT with multiple additional settings recommended to me by the pros. For future references, I may call this modified set of parameters SGVAD (Schmidt-Gusev-Velasco-Ablett-Dart) after all involved: In this order of letters, just so it reminds "squad". kNodes/sec visibly improved, and it remains to be seen to which extent ELO tracks. I attribute most of the improvement to the PGO (Profile-Guided Optimization) strongly recommended by Jim Ablett.

Yes, the "MT" compile works fine for me. Thanks. And no doubt, Ablett is "the man" when it comes to compiles.



Best,

george

[velmarin]

I think you are right on the edge considering it- and all it might take is a light push- to ask you to seriously consider getting back in the game with 2.3. It is just this engine, meaning 2.2 xTreme, can play at times like a world champion. And I don't care what Dmitri or anyone else does- they can't match what you can do. And that is not a criticism of him- or anyone else. He would agree with me as well.

For the record, yes, I totally agree with George! The road is open for you, Norm, to either release Fire 2.3, or join Firenzina GitHub and work on Firenzina 2.3, also with you as the primary author, as it should be. Assuming that you get along with Jose Velasco who gave the advice on how one could make Fire stronger (http://www.talkchess.com/forum/viewtopi ... 00&t=47152), I'd like to invite him to the team.

Friend Gusev,
I thank you for considering the advice.
But there is no merit on my part, do not have to give me any recognition,
My first compilation the Ippolit was with the advice given for Norman in Immortal Forum
in fact until recently I have read many times that thread.

The advice I gave you are very simple, nothing of merit.

By the way, one last tip, Intel compiler 13 makes a highly optimized code, with the profiles you will not advance anything.
Make a universal build and forget Kns get more, that's not going to give ELO program.
Start working on the code, and put their ideas and start testing parameters.
Really skip the compilation.

[Gusev]

The advice I gave you are very simple, nothing of merit.

Simple, but it worked! Thanks!!

By the way, one last tip, Intel compiler 13 makes a highly optimized code, with the profiles you will not advance anything.
Make a universal build and forget Kns get more, that's not going to give ELO program.

I believe that you're right, and with some help from the pros, I may be very near the point where any further games with compiler settings won't pay off. At that point, I will proceed to make, test, and release the requested non-Intel-specific, non-SSE4.2, non-popcnt, and x32 compiles, just as you had also advised.