Rybka 4 Matches

[JackStraw]

Here's a compilation of my R4 tests so far. All games at 4 min / + 2 sec , on a i7-940 @ 3.75 , 3 cores ea , ponder off , 512 mb hash , no egtb , klo_50 openings , Vista x64 , Home-Premium edition , No large pages , Fritz 11 gui , all x64 engines ,

Rybka 4 SSE42 ............+30/=59/-11.........59.5/100
Stockfish 1.7.1............+11/=59/-30.........40.5/100

Rybka 4 SSE42.............+26/=52/-22.........52/100
IvanHoe 63.2s.............+22/=52/-26.........48/100

Rybka 4 SSE42.............+19/=63/-18.........50.5/100
IvanHoe B63Mod12c.......+18/=63/-19........49.5/100
(intel SSE4.2 version)

Rybka 4 SSE42..............+21/=64/-15........53/100
Fire 1.2......................+15/=64/-21........47/100

Rybka 4 x64.................+17/=56/-27........45/100
Fire 1.2......................+27/=56/-27........55/100

Rybka 4 SSE42..............+25/=46/-29........48/100
Fire 1.3 popcnt.............+29/=46/-25........52/100

Rybka 4 SSE42..............+15/=68/-17........49/100
Houdini 1.01................+17/=68/=15........51/100
(Popcnt vers)

Also , was testing Fire 1.1 with Dr. Wael Deeb setttings vs Rybka 4 last night , but comp rebooted after windows updates ........ will get back to that ........ Fire was leading after 54 games +10/=36/-8

Testing IvanHoe T63B x64 now........ after 31 games R4 leads +10/=16/-5

Last 2 tests , using Robbo bases + 3-4-5 egtb for R4

Tom

[Charles]

Assuming the Fire 1.3 you tested was with default settings?

Very surprised to see R4 lose to Fire though it is quite close ..

[JackStraw]

Used default settings , except upped pawn hash to 64 .........

Small samples / high margin of error ......... prob if I ran the same 100 game test , the outcome might easily swing 4-5 pts the other way.......

About all I conclude is that R4 is about the same approximate basic strength as Houdini , Fire , the top ivanHoe's , and that on a given day , given game , anything can happen...............

[Tomcass]

Gui: Fritz 12
Time control: 4min + 2 sec.
Book: HS Masterbook
Ponder: Off
No Triple or Robbo Bases
Quad + Hyperthread i7 975 12 M Ram.
DamirsRK 42 used ONE CORE and 32w.
DR4 Ivan and Houdini used EIGHT THREADS and 64w.(Only as a reference, in terms of kns Ivan reached a speed about 700% that of DamirRK)

201006Houdinix8-2 2010

1 2 3 4
1 Deep Rybka 4 x64 11.5 - 8.58.5 - 11.515.5 - 4.5** 35.5/60
2 IvanHoe-T63Mini10 8.5 - 11.512.5 - 7.513.5 - 6.5 ** 34.5/60
3 Houdini DEVEL x64 8_CPU 11.5 - 8.57.5 - 12.513.0 - 7.0 ** 32.0/60
4 DamirsRybkaKiller42_w32 4.5 - 15.56.5 - 13.57.0 - 13.0 ** 18.0/60

Again a good result for DRK42 against three very strong engines, despite the huge speed handicap.

Tom.

[JackStraw]

More tests , same conditions as above , except these using RobboBases , and 3-4-5 egtb for Rybka
Also HT turned off on all tests........


Rybka 4 SSE42................+33/=52/-15................59/100
IvanHoe T63B.................+15/=52/-33................41/100

Rybka 4 SSE42................+27/=55/-18................54.5/100
IvanHoe T63B.................+18/=55/-27................45.5/100
(SSE4.2x64 vers)


Rybka 4 SSE42................+23/=53/-24................49.5/100
Fire 1.1 PopCnt...............+24/=53/-23................50.5/100
( using Dr. Wael Deeb settings )

JackStraw/Wichita

[Dr.Wael Deeb]

More tests , same conditions as above , except these using RobboBases , and 3-4-5 egtb for Rybka
Also HT turned off on all tests........


Rybka 4 SSE42................+33/=52/-15................59/100
IvanHoe T63B.................+15/=52/-33................41/100

Rybka 4 SSE42................+27/=55/-18................54.5/100
IvanHoe T63B.................+18/=55/-27................45.5/100
(SSE4.2x64 vers)


Rybka 4 SSE42................+23/=53/-24................49.5/100
Fire 1.1 PopCnt...............+24/=53/-23................50.5/100
( using Dr. Wael Deeb settings )

JackStraw/Wichita

Hi,
Thanks for testing my settings and yet again confirming that this personality of Fire 1.1 is extremely strong....
Cheers,
Dr.D

[Taner Altinsoy]

More tests , same conditions as above , except these using RobboBases , and 3-4-5 egtb for Rybka
Also HT turned off on all tests........


Rybka 4 SSE42................+33/=52/-15................59/100
IvanHoe T63B.................+15/=52/-33................41/100

Rybka 4 SSE42................+27/=55/-18................54.5/100
IvanHoe T63B.................+18/=55/-27................45.5/100
(SSE4.2x64 vers)


Rybka 4 SSE42................+23/=53/-24................49.5/100
Fire 1.1 PopCnt...............+24/=53/-23................50.5/100
( using Dr. Wael Deeb settings )

JackStraw/Wichita

Hi,
Thanks for testing my settings and yet again confirming that this personality of Fire 1.1 is extremely strong....
Cheers,
Dr.D

I'm also testing your settings in 2 different tournaments. On my laptop with celeron 1400, 32 bit single cpu and 2'2" time control its not that strong (about same as Fire 1.2 but worse than 1.3 and robbolito, deep rybka and houdini 1 cpu), this tournament consists of 300 gmes played among 16 diff engines and 2600 games played so far and it took me almost 2 weeks playing 7/24. But on my desktop x2 amd 4200 64 bit, 3'3" tournament it's in top 3 but still many games to be played. In this one only 360/1360 games played. Will report on both when completed.