testing R4 against Fire 1.31 on 8 cpu at 4+2

[Guenther]

here is a game where Fire thinks its fine and Rybka says mate in 39.

[Event "0000-7160E6B413, Blitz:4'+2""]
[Site "0000"]
[Date "2010.06.06"]
[Round "27"]
[White "Deep Rybka 4 x64"]
[Black "Fire 1.31 x64"]
[Result "1-0"]
[ECO "B92"]
[Annotator "0.04;-0.01"]
[PlyCount "193"]
[TimeControl "240+2"]

...snip...} 59. a7 {5.00/12 1} Kb6 {0.63/24 1 (Ra2+)
} 60. Rb8+ {#39/19 8} Kxa7 {0.63/22 0} 61. Rb5 {#37/0 0} Ra2+ {#36/0 0} 62. Kf3
{#36/0 0} Ra1 {#35/0 0} 63. Rxg5 {#35/0 0} Rf1+ {#34/0 0} 64. Kg2 {#34/0 0} Rb1
{#33/0 0} 65. Rd5 {#33/0 0} Kb6 {#32/0 0} 66. g5 {#32/0 0} Kc6 {#31/0 0} 67.
Rd3 {#31/0 0} Re1 {#30/0 0} 68. g6 {#30/0 0} Re8 {#29/0 0} 69. Kh3 {#29/0 0}
Rg8 {#28/0 0} 70. Rg3 {#28/0 0} Kd6 {#27/0 0} 71. Kh4 {#27/0 0} Ke7 {#26/0 0}
72. Kg5 {#26/0 0} Ra8 {#25/0 0} 73. Kh6 {#25/0 0} Rh8+ {#24/0 0} 74. Kg7 {#24/
0 0} Rh1 {#23/0 0} 75. Re3+ {#23/0 0} Kd6 {#22/0 0} 76. Kg8 {#22/0 0} Kd5 {#21/
0 0} 77. g7 {#21/0 0} Kd6 {#20/0 0} 78. Re4 {#20/0 0} Rh3 {#19/0 0} 79. Kf7 {
#19/0 0} Rf3+ {#18/0 0} 80. Kg6 {#18/0 0} Rc3 {#17/0 0} 81. Rg4 {#17/0 0} Rc8 {
#16/0 0} 82. Kf6 {#16/0 0} Rg8 {#15/0 0} 83. Rg5 {#15/0 0} Re8 {#14/0 0} 84.
g8=Q {#14/0 0} Rxg8 {#13/0 0} 85. Rxg8 {#13/0 0} Kc5 {#12/0 0} 86. Ke5 {#12/0 0
} Kc4 {#11/0 0} 87. Rd8 {#11/0 0} Kc3 {#10/0 0} 88. Rd4 {#10/0 0} Kb3 {#9/0 0}
89. Kd5 {#9/0 0} Kc3 {#8/0 0} 90. Kc5 {#8/0 0} Kb3 {#7/0 0} 91. Rc4 {#7/0 0}
Ka3 {#6/0 0} 92. Rb4 {#6/0 0} Ka2 {#5/0 0} 93. Rb5 {#5/0 0} Ka1 {#4/0 0} 94.
Kd4 {#4/0 0} Ka2 {#3/0 0} 95. Kc3 {#3/0 0} Ka1 {#2/0 0} 96. Kc2 {#2/0 0} Ka2 {
#1/0 0} 97. Ra5# {#1/0 0} 1-0

No good example. It only shows that Rybka used 6men tablebases
in search(as it should) and one ply later they kicked in for both from the GUI used.

Guenther

[Dr.Wael Deeb]

All in all,it seems like Rybka 4 is stronger than Fire 1.31....But Houdini is a different story indeed....
Dr.D

On big hardware like this, R4 is definitely stronger than Fire, and I'd guess, Houdini. On littler hardware it is much different.

Peter

Agreed Peter....it looks like on octal machines things are a little bit different....if we assume that Rybka 4 is a dumped version of the cluster Rybka,then there is nothing more to say....the scaling on the bigger hardware will show the difference....
Dr.D

Fire 1.3 scaling is quite good.

didn't test it yet....in my waiting list though....
Dr.D

[ArmyBridge]

settings are.out of box.first 110 games book Amazing(based on high quality corr games) last 60 HS MasterBook 2.ponder off,win64 turbo 3 tbs.

Final update

Rybka 4-Fire 1.31

+62=91-17 (170 games) +94 ELO.

to many exceptions to play more,all pairs of games when an exception occured were deleated.

i will post a few of the games if you wish.

Hi Paul, Interesting... and yes DR4 in octal seems to be stronger than everything, but I wish suggest to u that try with this:
IvanHoe-T_Ghost4
I´m pretty sure that this IH will beat DR4 in 8 cores, if u run the test, set TC to x' + y'', IvanHoe-T_Ghost4 have no good time handling, and lost a lot of games in time if you don't put incremental time.
PD colud u upload the DR4-FB1.31 match?? tnx in advance!!

[frcha]

Houdini_w32_4CPU - Rybka 3_(cont 0) : 67.5/100 44-9-47 (11=01==1=1====00=11=0=====1=1===111111=11=1=0=1=1=1===1=111=11111===11==1==011=0==1=1===11==1=111010) 68% +131
Houdini_w32_4CPU - Rybka 4 w32 : 46.5/100 22-29-49 (===0==011=1===1=0==010==00====00001001=0=0=1=0001===101=01==011=1=111====1010=00=0=====0===0===0=11=) 47% -21

4 cpu ponder=off 32 bit 0+3s tc silver suite

2 points ...
R4 wins by small margin .
Houdini beats R3 by incredible 68%

[NATIONAL12]

perhaps Norman would like to chip in

yes ..clearly Rybka 4 is far superior to Fire...!?
you can't actually see me (without a webcam) but: i'm actually on the floor...prostrating myself in complete humbleness to you and Rybka 4's Octal magnificence!

you were right all along Paul...!
nobody will ever again doubt your incredible chess engine genius!

with that said, i humbly beg forgiveness....
Rybka 4 is awesome !!!
clearly much stronger than anything else!
(Paul Watson's testing guarantees it!)

buy it today!
(how much is it)?

PS-
pehaps if you contact Vas with these results and an indication of your dedicated and fervent fanatacism...you and he can become frends!?

i will ignore all your sarcasm.

i am testing now on a quad at 8+4 and despite 4 exceptions by Fire i have over 70 games with atm +76 ELO for Rybka.

if you want to do something sensible stop Fire crashing.

[M ANSARI]

R4 on 8 cores is a completely different beast and have had almost identical results. Actually R3 on 8 cores was also extremely strong, but R4 seems to have even better scaling. Now if you factor in the fact that you still have Large Pages to add and maybe SSE42 for a little more, then this can only become even more of a factor. In my tests on 8 cores R4 had scores of over +100 ELO against Houdini and +90 ELO against FB 1.31 and was scoring in around 73% to 75% against Naum 4.2 and Stockfish 1.71. These results are probably even better as there seems to be quite a few number of games that R4 loses on time in drawn or even won positions. Unfortunately this large superiority does not hold up on less cores or smaller hardware. To be honest I still can't figure that one out as the difference seems incredibly huge.

[ArmyBridge]

Hi, seems that R4 on 8 cores beat everithing....... almost, here it is:
" Blitz 3'+0" no one IvanHoe compile has won against Deep Rybka 4.
No Peterpan's super versions,no Ahmed's Mini's.
A old poor compile without parameters and without Robbo Bases with name
T Ghost 2 x64 beat the great fish!!!
Intel Xeon X5355 (8x2660 MHz),Blitz 3'+0",Hash 128,Ponder off,GUI 12,
book: HS-7moves.ctg:
T Ghost 2 x64 - Deep Rybka 4 : 26.5 -23.5 . 53% - 47%.
(+12 -9 =29)
For me this is not a surprise,Ghost 2 has won clearly since many weeks also
against Rybka 3 and was in my rating list everytime one of the best compiles to 8 real cores.I don't know how he performs to 2 and 4 cores but to 8 threads his scaling is super!"
Whata!!! IH T Ghost 2 x 64 the "MATA RYBKA"

Regards!!

[NATIONAL12]

i am also testing on a quad at 8+4,not exactly micky mouse blitz like above post.

i have 200 games atm,this takes time which most posters dont seem to have.they seem to want instant gratingaftion almost like an orgasm.

results are not as good as 4+2 on 8 cpu(please nobody mix core with threads).atm +47 ELO but this includes (6 exceptions.ie crashes)

will post in due course

[gerold]

i am also testing on a quad at 8+4,not exactly micky mouse blitz like above post.

i have 200 games atm,this takes time which most posters dont seem to have.they seem to want instant gratingaftion almost like an orgasm.

results are not as good as 4+2 on 8 cpu(please nobody mix core with threads).atm +47 ELO but this includes (6 exceptions.ie crashes)

will post in due course

Looks like Rybka is geared for 8+cpu. Wow. Just think what it
will do on cluster computer setup.

[NATIONAL12]

i am also testing on a quad at 8+4,not exactly micky mouse blitz like above post.

i have 200 games atm,this takes time which most posters dont seem to have.they seem to want instant gratingaftion almost like an orgasm.

results are not as good as 4+2 on 8 cpu(please nobody mix core with threads).atm +47 ELO but this includes (6 exceptions.ie crashes)

will post in due course

Looks like Rybka is geared for 8+cpu. Wow. Just think what it
will do on cluster computer setup.

my final results are available on Rybka forum after 10 exceptions with Fire 1.3.in which in 8 of these games it crashed when evals against it were over 4. 230 games.