Not logged inRybka Chess Community Forum
Forum RybkaChess Help Search Register Login
Up Topic The Rybka Lounge / Computer Chess / SCCT - Deep Fritz 13 w32 6c / Houdini 2.0z x64 6c
- - By Sedat Canbaz (*****) Date 2012-06-06 08:47
Hello Chess Friends,

I am pleased to announce:
- Very soon, Deep Fritz 13 and Houdini 2.0z will be started testing under SCCT:
http://www.sedatcanbaz.com/chess/scct-rating/

Some Notes:
-Houdini 2.0z is a new modified version by Salvador,for more details:
http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=24987
-Deep Fritz 13 is the latest release by ChessBase (Frans Morsch/Mathias Feist):
http://www.chessbase-shop.com/en/products/deep_fritz_13_download
-I hope to publish the current engine results in 10 days period of time

Best Wishes,
Sedat Canbaz
Parent - - By Carl Bicknell (*****) Date 2012-06-06 16:19
when are you starting? On a single core I'm only expecting 20-30 elo improvement on DF13 c/p with F13.
Parent - - By Sedat Canbaz (*****) Date 2012-06-06 17:53

>when are you starting? On a single core I'm only expecting 20-30 elo improvement on DF13 c/p with F13.


Right now i have started testing Deep Fritz 13 6c and Houdini 2.0z x64 6c
And i hope to publish the first results after 2-3 days

Best,
Sedat
Parent - - By Carl Bicknell (*****) Date 2012-06-06 19:10
I've just bought DF13 (for the GUI) - the tactical results are dire.
Parent - - By rocket (****) Date 2012-06-06 20:58
Is it not even better than "Fritz 13"? if you compare the two on 1cpu?
Parent - By Carl Bicknell (*****) Date 2012-06-06 21:20

> Is it not even better than "Fritz 13"? if you compare the two on 1cpu?


Given how slow the TWELVE core version plodded through the tactics suite, testing on ONE core would tie up my machine for many hours!

I'll post results soon
Parent - - By Mark Eldridge (****) Date 2012-06-06 21:17
If the tactical results are dire is it worth buying just for the GUI?
Parent - By Carl Bicknell (*****) Date 2012-06-06 21:34

> If the tactical results are dire is it worth buying just for the GUI?


2 considerations:

i) do you own machines or want to rent machines that are fast when you are out and about using your laptop? If so then maybe yes.

ii) Is the Rybka Cluster going to be available to rent? Lukas has said 'yes' somewhere else, but there are no details yet, so maybe it'll come later this year, perhaps.
Parent - - By Carl Bicknell (*****) Date 2012-06-06 21:37
Basically the GUI is exactly the same as F13, but with a cloud button as well.
Parent - By Mark Eldridge (****) Date 2012-06-06 21:43
Thank you for the information. I don't have F13 my last Fritz was D11
Parent - By tomgdrums (****) Date 2012-06-09 13:34
So how does the cloud button work?  Do you have to pay every time you want to use "cloud" engines?  Or are they just there in the cloud for use?

How is single core Fritz 13 for analysis?  Does it give any original analysis (especially in endgames) over different engines?  I am always on the look out for different engine opinions when going over my games but have only owned Fritz 10 but have lost the disk. 

Thanks!
Parent - - By Barnard (Bronze) Date 2012-06-06 20:54

>when are you starting? On a single core I'm only expecting 20-30 elo improvement on DF13 c/p with F13.


just one question my friend;you expect +20 or 30 elo improvement running DeepFritz 13 vs Fritz 13 on a single core?im right about that?
Parent - - By Carl Bicknell (*****) Date 2012-06-06 21:31

> you expect +20 or 30 elo improvement running DeepFritz 13 vs Fritz 13 on a single core?im right about that?


Well that's my GUESS, yes.

Here are some early results. Bear in mind the game sample is very small and the time control fast:
G1+1
20 game Nunn suite

Deep Fritz 13 (12 core) vs Deep Fritz 12 (12 Core)
11.5 - 8.5
+53 elo
Parent - By Barnard (Bronze) Date 2012-06-06 21:35
ah,ok...sorry,from your post i thought you are running DeepFritz oon a single core,and that has no sense,and that is why i asked

but yes,20 games is a very small sample;just one example,when i ran the match of 1000 games between the tuned Houdini2.0z and Critter,with near 100 games played,it showed an improvement of more than 90 elo points,but after the rest of the match (900 more games),the elo dropped to +57
Parent - - By Barnard (Bronze) Date 2012-06-06 19:18
Hi

Sedat,thanks for your help testing the tuned version Houdini2.0z

now im finishing a 1000 games match between Houdini2.0z and Critter 1.4,and after 976 games,the tuned version shows a +56 elo points over Critter

regards

Salvador
Parent - - By Sedat Canbaz (*****) Date 2012-06-07 17:38 Edited 2012-06-07 18:00
Dear Salvador,

>Sedat,thanks for your help testing the tuned version Houdini2.0z


Not at all...

Hmm...honestly i am disappointed by the current performance of Houdini 2.0z x64 6c

It seems Houdini 2.0z modified version does not suit/not optimized for SCCT conditions

Unfortunately,the current Houdini 2.0z results are 78.5/139 (56% winning percentage) that means approx. 130 Elo weaker than Houdini 2.0c

And i am not sure to continue or to cancel the Houdini 2.0z test  (in case of similar performance up to 200 games,probably then i will stop the test)

I know too that we need more games,but i highly doubt even after 1000 games that Houdini 2.0z will be performed better than Houdini 2.0c (default settings)

Best,
Sedat
Parent - By Barnard (Bronze) Date 2012-06-07 20:21
Hi

i cant understand that Sedat...from all the test,those test shows an improvement of around + 13 Houdini 2.0z over Houdini 2.0c,and also ran test Houdini2.0z against Critter (+57 for Houdini2.0z) and against Komodo4 (1000 games test,and after 231 games,it shows a +66 for Houdini2.0z),so i really cant understand what happens...if you told me that the tuned version is equal at strength over Houdini 2.0c,or just 4 points stronger,i will understand...but 13o elo points weaker than the Houdini2.0c,when it shows an improvement of +13 against Houdini2.0c,is something that i cant understand why....

but thanks for your help Sedat
Parent - By Barnard (Bronze) Date 2012-06-07 21:35
look that partial results that i received (the screen capture)a few minutes ago Sedat
Parent - - By Barnard (Bronze) Date 2012-06-08 01:07
Sedat,i found the problem and why those values doesnt work...pm me when you read that,and also i sent you a pm trying to explain

Regards

Salvador
Parent - - By free Willy (*) Date 2012-06-08 19:28
Bernard Ok, I'm sorry for you. So it's probably not so easy to make a top engine even more. But luckily we have Mr. Houdart with Houdini3. The thing is probably over and now it is just embarrassing. The whole-importance has fortunately an end. And now no excuses why it did not work. Sorry
Parent - By Barnard (Bronze) Date 2012-06-08 20:38 Edited 2012-06-08 20:50
free willy,i know where the problem was,and why the Houdini 2.0z performed so wrong with Sedat tests

here are 2 screen captures of running tests against Komodo 4 (both engines running with 1 thread) and against Stockfish

p.s.i will upload a screen capture of the match im running on my computer when the match will be at 250 games,now is 230 games,and i will try to upload screen captures at 500 and 750 games,and of course,the final result
Parent - - By Sedat Canbaz (*****) Date 2012-06-08 16:13 Edited 2012-06-08 16:33
Dear Friends,

Last Updated Engine:
Deep Fritz 13 w32 6c

SCCT Rating List:
http://www.sedatcanbaz.com/chess/scct-rating/

Some Notes:
-Deep Fritz 13 performed just 12 Elo stronger than Deep Fritz 12
*I hope, Deep Fritz 13 will be ranked at higher places (after more games...)
-Houdini 2.0z test is cancelled,due to there is something wrong with its strenght
-The SCCT Rating List includes ‘No double games’ and ‘No games up to 15 moves’

Best Wishes,
Sedat
Parent - - By Sedat Canbaz (*****) Date 2012-06-09 10:59 Edited 2012-06-09 11:11
Just to make it more clear,
Houdini 2.0z's low performance is not due to my fault or its not due to Salvador's fault
Note that I used exactly the same modified Houdini 2.0z exe (with the recommended settings by Salvador)

In my opinion,it's just a Houdini engine strength issue,its probably played weaker due to HoudiniConfig is not used for each PCs
To be honest,we thought (Salvador and Me) that its enough one time modifying ...
In other words,there was only one time modifying on other PC (not in my i7 6 core machines),but it seems (for maximum performance) HoudiniConfig should be used for each PC

But however, i dont want to promise,maybe when i will have free time,i can re-start again the Houdini 2.0z test

Best,
Sedat
Parent - By Sedat Canbaz (*****) Date 2012-06-11 23:36
After 820 games,Deep Fritz 13's performed 30 Elo higher than Deep Fritz 12

For SCCT Rating List:
http://www.sedatcanbaz.com/chess/scct-rating/

Kind Regards,
Sedat
Parent - - By Geomusic (*****) Date 2012-06-09 19:19 Edited 2012-06-09 19:23
Ok, -so let me get this straight-  I know I just crawled out from under a rock with this question, but with the new DF 13 it is my understanding that you can make your own local chess cluster without expensive or intimate knowledge of  protocols, batch files or OS's infact you just download DF13 on each machine, connect them to a lan and you are set?!! This might be the best thing invented since the mouse wheel! So easy to make a beast machine now!!
Though because of this simplicity the next freestyle might be a stalemate(har har)!
Parent - By Labyrinth (*****) Date 2012-06-09 23:59
I don't see how they handle the latency issue between computers across the globe, but supposing they have a working solution you could have a 'beast' machine indeed :-)

Like I was saying, global supercomputer chess tournament :-P
Parent - - By Pia (****) Date 2012-06-10 02:55
Improved version of Houdini 2 would make fun of Fritz, indeed 
Blitz:1', Pondering 2 out of 4 cores each, Perfect 2012 book
                        
1   Houdini 1.5a w32    +16/=3/-1 87.50%   17.5/20 (contempt = 2)
2   Deep Fritz 13       +1/=3/-16 12.50%    2.5/20

+/- 338 Elo
Parent - - By Barnard (Bronze) Date 2012-06-10 03:11
you used 1 minute without increment?

just asking
Parent - - By Pia (****) Date 2012-06-10 03:42
True. Now look at 4 min + 2 sec increment but using 1 core pondering. Same -338 Elo but games are yet only so few 
Blitz:4'+2"

                        1234
1   Houdini 1.5a w32    111½    3.5/4
2   Deep Fritz 13       000½    0.5/4
Parent - - By Barnard (Bronze) Date 2012-06-10 07:43
same ratio :smile:
Parent - - By Pia (****) Date 2012-06-10 14:07
Continued. <> 191 Elo 
Blitz:4'+2"
                        
1   Houdini 1.5a w32    +40/=22/-6  75.00%   51.0/68
2   Deep Fritz 13        +6/=22/-40 25.00%   17.0/68
Parent - By Barnard (Bronze) Date 2012-06-10 20:47
now the elo is more 'real',and as soon you will play more games,it will stabilize the elo
- - By rocket (****) Date 2012-06-08 03:15 Edited 2012-06-08 03:25
So if Deep fritz 13 is not stronger tactically then what exactly is improved compared to version 12? The rating difference is quite large(around 40-50 points). Does it reach higher depths faster?
Parent - By Pia (****) Date 2012-06-10 00:40

>Does it reach higher depths faster?


New game

Analysis by Deep Fritz 12:

1.d4 Nf6 2.Nc3 d5 3.Bf4 g6 4.e3 Bg7 5.Be2 0-0 6.h4 Nc6 7.h5 gxh5 8.Nf3 Bf5 9.Nh4
  +/=  (0.30)   Depth: 21/50   00:01:08  394mN

Analysis by Deep Fritz 13:

1.d4 Nf6 2.c4 e6 3.Nc3 d5 4.Nf3 Nc6 5.e3 Be7 6.Bd2 0-0 7.Be2 h6 8.0-0 dxc4 9.h3 b6
  +/=  (0.27)   Depth: 24/48   00:01:28  484mN
Parent - - By Geomusic (*****) Date 2012-06-10 06:28
I would buy DF13 4x over just for the ability to connect several computers together. This is so rad!
Parent - By Geomusic (*****) Date 2012-06-10 06:29
H2C is almost at the mystical 3400 ELO barrier on SCCT :)
- By dragon49 (****) Date 2012-06-10 03:44
Did anybody expect a different result?
Up Topic The Rybka Lounge / Computer Chess / SCCT - Deep Fritz 13 w32 6c / Houdini 2.0z x64 6c

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill