Strelka 5.5 x32 is released....

[Houdini]

Code: Select all

Games Completed = 850 of 10000 (Avg game length = 7.112 sec)
Settings = RR/16MB/50ms per move/M 700000cp for 1000 moves, D 3000 moves/PGN:C:\Users\Downloads\LittleBlitzer\swcr.pgn(5120)
Time = 10402 sec elapsed, 111972 sec remaining
 1.  Houdini 1.5a w32         	370.0/850	202-312-336  	(L: m=312 t=0 i=0 a=0)	(D: r=254 i=41 f=40 s=1 a=0)	(tpm=47.4 d=11.4 nps=1076409)
 2.  Strelka 5.5              	480.0/850	312-202-336  	(L: m=202 t=0 i=0 a=0)	(D: r=254 i=41 f=40 s=1 a=0)	(tpm=55.5 d=12.1 nps=968204)

Code: Select all

    Program                            Score     %     Elo    +   -    Draws

  1 Strelka 5.5                    : 480.0/850  56.5   3223   18  18   39.5 %
  2 Houdini 1.5a w32               : 370.0/850  43.5   3177   18  18   39.5 %

Nobody seems to be able to reproduce this result at more normal time controls, can you?
I wasted some CPU time to play an 8000-game match between Houdini 1.5a and Strelka 5.5 at time control 8"+0.08". Houdini 1.5a won:

Code: Select all

   1 Houdini 1.5a w32   12 8000.0 (4280.5 : 3719.5)
                           8000.0 (4280.5 : 3719.5)
   2 Strelka 5.5       -12 8000.0 (3719.5 : 4280.5)
                           8000.0 (3719.5 : 4280.5)

(note: this was played on hardware with SSE4.2, giving Houdini 1.5 a small speed advantage - at this TC worth about 5 to 10 Elo)

In analysis Strelka 5.5 and Strelka 5.1 produce nearly identical results - very often even identical node counts - the changes between the two versions must be minimal.
As far as I can see, Strelka 5.5 is still 99% search and eval of Houdini 1.5, with some other parts (e.g. the time management) taken from the Ivanhoe source.
In view of that, the reactions in this thread are hilarious...

Robert

[gleperlier]

ouch ! This post by Robert will calm down many people

Gab

[Houdini]

ouch ! This post by Robert will calm down many people

Gab

I doubt it, I've told nothing new...
People only hear what they want to hear.

Robert

[ATOMICC]

Hello,

Strelka doesn't work on Fritz13 GUI ? Is it UCI ?

Gab

On my "Fritz 13 GUI" Strelka 5.5 UCI" works just fine!

But will it do a dozen games with Ponder=On without causing exception errors? Has that been fixed yet?

[gleperlier]

Hello,

Strelka doesn't work on Fritz13 GUI ? Is it UCI ?

Gab

On my "Fritz 13 GUI" Strelka 5.5 UCI" works just fine!

But will it do a dozen games with Ponder=On without causing exception errors? Has that been fixed yet?

No problem with Ponder On here.

Gab

[ATOMICC]

Hello,

Strelka doesn't work on Fritz13 GUI ? Is it UCI ?

Gab

On my "Fritz 13 GUI" Strelka 5.5 UCI" works just fine!

But will it do a dozen games with Ponder=On without causing exception errors? Has that been fixed yet?

No problem with Ponder On here.

Gab

I'm sorry. I should have clarified. Will it run games with Ponder=On using the Fritz GUI?

[ATOMICC]

...
In view of that, the reactions in this thread are hilarious...

Robert

"Where have you gone, Joe DiMaggio?
A nation turns it's lonely eyes to you (Woo, woo, woo)"

[gleperlier]

Hello,

Strelka doesn't work on Fritz13 GUI ? Is it UCI ?


I'm sorry. I should have clarified. Will it run games with Ponder=On using the Fritz GUI?

Yes

[ATOMICC]

Hello,

Strelka doesn't work on Fritz13 GUI ? Is it UCI ?


I'm sorry. I should have clarified. Will it run games with Ponder=On using the Fritz GUI?

Yes

Thank you, Gabriel!

[Laskos]

Nobody seems to be able to reproduce this result at more normal time controls, can you?
I wasted some CPU time to play an 8000-game match between Houdini 1.5a and Strelka 5.5 at time control 8"+0.08". Houdini 1.5a won:

Code: Select all

   1 Houdini 1.5a w32   12 8000.0 (4280.5 : 3719.5)
                           8000.0 (4280.5 : 3719.5)
   2 Strelka 5.5       -12 8000.0 (3719.5 : 4280.5)
                           8000.0 (3719.5 : 4280.5)

(note: this was played on hardware with SSE4.2, giving Houdini 1.5 a small speed advantage - at this TC worth about 5 to 10 Elo)

In analysis Strelka 5.5 and Strelka 5.1 produce nearly identical results - very often even identical node counts - the changes between the two versions must be minimal.
As far as I can see, Strelka 5.5 is still 99% search and eval of Houdini 1.5, with some other parts (e.g. the time management) taken from the Ivanhoe source.
In view of that, the reactions in this thread are hilarious...

Robert

At move/0.05s the result of 2,000 games match is:

Code: Select all

    Program                            Score     %       Elo    +   -    Draws

  1 Strelka 5.5                    : 1145.5/2000  57.3   3225   12  12   38.6 %
  2 Houdini 1.5a w32               :  854.5/2000  42.7   3175   12  12   38.6 %

+51 Elo points for Strelka. Also, for Arasan14 test suite (5s+1ply / position) the results are so much in favour of Strelka 5.5 that I am suspecting some oddity. Maybe I have to restart my comp and check for GUI bugs and such, but as of now Strelka seems clearly stronger.

Kai