Strelka 5 by Jury Osipov is out!

[ThinkingALot]

The point is, that this x64 JA compile is probably just a w32 source compiled to x64 (that's my guess based on the performance difference between w32 and x64): no native x64 lsb(), no hardware popcnt, 32-bit optimized magics, no large pages... And the question is, how damn strong Strelka is going to be with all these features turned on? I think the answer is clear: it will share the "Top 1-2" place.
Another question: how strong will it be when Jury tunes evaluation parameters a bit?

[thorstenczub]

[Event "Arena Turnier"]
[Site "ORION8-LATITUDE"]
[Date "2011.10.18"]
[Round "1"]
[White "ProDeo1.6 [mach011]"]
[Black "Strelka-5-64"]
[Result "0-1"]
[BlackElo "2200"]
[ECO "A90"]
[Opening "Holländische Verteidigung"]
[Time "18:13:17"]
[Variation "2.c4 Nf6 3.g3 e6 5.Nf3 d5 6.Qc2 Bd6"]
[WhiteElo "2200"]
[TimeControl "120+2"]
[Termination "adjudication"]
[PlyCount "87"]
[WhiteType "program"]
[BlackType "program"]

1. c4 {0.00/1 0} e6 2. Nf3 {0.00/1 0} f5 3. d4 {+0.30/10 5} Nf6 4. g3
{0.00/1 0} d5 5. Bg2 {0.00/1 0} c6 6. O-O {0.00/1 0} Bd6 7. Nc3 {0.00/1 0}
O-O 8. Qc2 {0.00/1 0} Ne4 9. Rb1 {0.00/1 0} Qe7 10. b4 {0.00/1 0} Bd7 11.
b5 {0.00/1 0} Nxc3 12. Qxc3 {0.00/1 0} cxb5 13. cxb5 {0.00/1 0} Rc8 14. Qb3
{+0.17/11 7} a6 {+0.40/16 8} 15. Bg5 {+0.32/10 6} Qe8 {+0.70/15 3} 16. bxa6
{+0.17/10 7} Rxa6 {+0.76/15 4} 17. Ne5 {+0.11/10 4} Bxe5 {+1.39/13 4} 18.
dxe5 {+0.32/5 0} Nc6 {+1.19/16 7} 19. Qb2 {+0.09/11 7} Na5 {+1.75/16 5} 20.
Bf3 {-0.35/10 3} Bc6 {+2.38/15 3} 21. Rbc1 {-0.93/11 6} Nc4 {+2.53/16 5}
22. Qa1 {-1.04/11 4} h6 {+2.91/17 3} 23. Bf4 {-1.32/13 7} g5 {+3.17/17 4}
24. Be3 {-1.40/12 4} Nxe3 {+3.12/18 8} 25. fxe3 {-1.02/5 0} Rca8 {+3.24/18
16} 26. Rc2 {-1.39/12 4} Ra3 {+3.24/18 17} 27. Bg2 {-1.39/11 9} Rxe3
{+3.66/16 9} 28. Rf3 {-1.34/11 4} Re4 {+3.81/16 4} 29. Rfc3 {-1.38/11 5}
Rb4 {+3.87/14 3} 30. a3 {-1.32/10 3} Qb8 {+4.41/16 6} 31. Qc1 {-1.66/10 4}
Qa7+ {+4.52/14 2} 32. e3 {-1.62/11 5} Ra4 {+4.52/13 0} 33. Ra2 {-1.66/10 4}
Qb8 {+4.33/14 2} 34. Rac2 {-1.64/11 4} Qxe5 {+4.90/15 6} 35. Rd3 {-1.73/11
5} h5 {+5.63/14 3} 36. Bf3 {-2.10/10 4} h4 {+6.29/14 5} 37. Rg2 {-2.10/9 2}
h3 {+6.59/15 3} 38. Ra2 {-2.27/11 4} Bb5 {+7.06/16 3} 39. Rc3 {-2.32/10 4}
Rc4 {+7.04/16 3} 40. Rxc4 {-2.34/11 3} Bxc4 {+7.04/13 0} 41. Rf2 {-2.36/11
2} b5 {+10.00/14 4} 42. Rc2 {-3.20/11 4} g4 {+10.19/14 3} 43. Bh1 {-3.22/11
2} Qd6 {+10.19/12 0} 44. Rc3 {-3.49/11 3} ... {0-1 Arena Entscheidung} 0-1

[alfons]

kingliveson;

does strelka ponder during matches under linux on your boxes?


I did setup a match between critter under xboard (Debian's testing* version: 4.5.3-1), and while critter does as befohlen, strelka refuses to do so (but it behaves otherwise during this match, so nothing gets aborted or so).

Of course strelka guesses the next move:
"< ENGINE bestmove d4e3 ponder e8g8"
but it just doesn't continue to calculate.

greetings of the fine season

*it's the OS I do use; 64-bit variant

[Jeremy Bernstein]

The point is, that this x64 JA compile is probably just a w32 source compiled to x64 (that's my guess based on the performance difference between w32 and x64): no native x64 lsb(), no hardware popcnt, 32-bit optimized magics, no large pages... And the question is, how damn strong Strelka is going to be with all these features turned on? I think the answer is clear: it will share the "Top 1-2" place.
Another question: how strong will it be when Jury tunes evaluation parameters a bit?

Well, it's no great feat to clone an engine and end up with something approximately the same strength as the original. Houdini 1.5a is also ~15 Elo weaker than Houdini 2. I don't want to pee in everyone's Wheaties, but who cares? Sure, it's nice to have a generic brand "Whodunnit" to compete with the name brand "Houdini", but they both came off of the same assembly line. OK, it plays a little differently, it obfuscates its analysis differently and whatever else, but it's kind of a joke. In my very personal, private, unpopular opinion.

Norman's points are 100% correct -- you can't kick Vas around, you can't kick the Ippofolks around, you can't kick Norman around and give Jury Osipov and/or Robert Houdart a hero's welcome. This community doesn't know what it wants. Strike that, it's a bunch of speed queens who will do anything, pay anything and tolerate anything to be able to say that the engine that thinks for them is stronger than the one that thinks for the other guy.

Flame on.

Jeremy

P.S.: if your only point is "yeah, but there's a Linux version," you can save it.

[kingliveson]

@ alfons,

I haven't tested pondering yet -- will later and post findings...



@ Jeremy,

I happen to agree that it's weird to see how IPPOLIT has been castigated when in-fact, we can trace its development from initial release to where it is today. I disagree however with the Jury/Houdart comparison. Jury, in this case is upfront in how Strelka 5 came about which he has admitted is from reversed engineered closed source RobboLito known as Houdini.

.
.
.
I don't know how you feel about it but at least in my opinion if there is one guy who deserves to get his code disassembled it is Mr Houdart.

.
.
.

[Prima]

I agree with Jeremy, & the quote from Azaad 100%.

Part of these clone-problems we see stems from customers: the people paying for something that is clearly, beyond doubt, proven to be a clone. Quite contrary to the "100 original" claimed by the author/s. Exactly the scenario I pointed out in 2009 or 2010; where it is feasible for crooked commercial authors to copy codes from GPL source codes, bundles the GPL codes as their original "hard work", and sell it to the [un]suspecting & useful public idiots.

Don't be surprise when everything from Stockfish and other strong GPLed source codes have been fully copied / stolen [because these thieves refuse to acknowledge GPL codes in their engines when the evidence proves otherwise] by some of these commercial crooks, their engine strength & "make-believe" progress stalls at, or equals to, the same strength level as Stockfish or RE engine. Then the tell-tale signs of 'stalled progress' becomes apparent, since these copy-cats HAVE TO, for once in their life, make use of their now torpid brains to be creative since there are no stronger GPL codes for these copy-cat-commercial-authors-wanna-be to steal from. Alternatively, these vultures can simply take the easy way out: Bail out. That is to say, run away. Vanish.

[thorstenczub]

Let's don't go there again (new approach - new paradigm)...

yes of course. of course the evaluations of strelka are a new aproach !
a very interesting one btw.

the way strelka kills very strong program because it has a different evaluation shows that thee is something interersting in this new method/function.

strelka nearly destroys hiarcs or houdini ...

[thorstenczub]

[Event "40/120"]
[Site "Atom netbook"]
[Date "2011.10.19"]
[Round "2"]
[White "Strelka 5"]
[Black "Houdini 2.0 w32"]
[ECO "B85"]
[Result "1-0"]

1. e4 {0s} c5 {book 0s} 2. Nf3 {15s} d6 {book 0s} 3. d4
{11s} cxd4 {book 0s} 4. Nxd4 {9s} Nf6 {book 0s} 5. Nc3
{12s} a6 {book 0s} 6. Be2 {10s (g3)} e6 {book 0s} 7. O-O
{9s} Be7 {book 0s} 8. f4 {11s} O-O {book 0s} 9. a4 {12s}
Nc6 {+0.02/20 8:00m} 10. Kh1 {31s (Be3)} Bd7 {+0.09/20
4:34m} 11. Be3 {2:18m} Rc8 {+0.06/20 7:17m} 12. Bf3 {32s
(Qd2)} Nb4 {-0.04/21 15:24m} 13. e5 {6:06m} Nfd5 {0.00/21
0s} 14. Nxd5 {2:31m} Nxd5 {-0.04/22 4:16m} 15. Bxd5 {8:24m}
exd5 {-0.04/22 0s} 16. Qf3 {2:47m} Qa5 {+0.03/20 6:49m}
17. c3 {8:17m (Rad1)} Rc4 {0.00/18 3:42m} 18. h3 {10:08m}
h6 {0.00/20 5:13m} 19. Bg1 {5:32m (Qg3)} Re8 {-0.05/19
3:47m} 20. Rae1 {15:53m (Rfe1)} Bf8 {-0.19/19 2:40m} 21. e6
{6:05m} fxe6 {-0.15/20 0s} 22. f5 {6:15m} e5 {-0.23/21 0s}
23. f6 {4:13m} Kh7 {-0.42/22 18:18m} 24. Qd3+ {1:35m
(fxg7)} e4 {-0.59/20 8:03m} 25. Qg3 {2:40m} g6 {-0.59/21
0s} 26. f7 {10:00m} Re5 {-0.88/21 0s} 27. Rf6 {18s} Rg5
{-0.74/18 3:06m} 28. Qxd6 {3s} Qd8 {-1.19/17 2:04m} 29. Ne6
{1:25m} Bxd6 {-1.71/18 1:49m} 30. Nxd8 {1:42m} Bf8
{-1.41/19 0s} 31. g4 {5:46m} Rc8 {-1.51/19 5:02m} 32. Ne6
{6:28m (Nxb7)} Re5 {-2.86/18 2:53m} 33. Nxf8+ {35s} Rxf8
{-3.93/21 1:26m} 34. Bc5 {3:17m} Kg7 {-4.28/22 2:53m}
35. Bxf8+ {1:57m (Rb6)} Kxf8 {-2.79/20 2:35m} 36. Rxg6
{5:17m} Kxf7 {-2.77/21 0s} 37. Rxh6 {47s} Bc6 {-2.94/20
4:11m} 38. b3 {6:12m} d4 {-3.11/21 0s} 39. cxd4 {1:00m} Rd5
{-3.10/22 2:48m} 40. Rd1 {3:37m} e3 {-3.12/20 0s} 41. Kg1
{3:34m} e2 {-3.18/21 5s} 42. Re1 {1:49m} 1-0

strelka handles houdini similar like it did with hiarcs.

[ernest]

Which exactly is the Strelka5.exe you are using?
The 17/10/2011 18:06 (292 352 bytes) ?

[stevenaaus]

Can anyone get Strelka Linux 32 bit to work in Scid's analysis window ?
I can't get PV happening. It goes great in ScidvsPC tournaments though.

All i get is a badly formatted info line ending with "pv (null)".