Ordo question

[Vinvin]

I also took the freedom to rename a few programs and marked all versions with less than 30 games and some 'others'...

1) For what I know :

Fritz 8 is not the same engine as Deep Fritz 8 with 1 CPU.
Fritz 9 is not the same engine as Deep Fritz 9 with 1 CPU.
....
same thing until Fritz 13.
For Fritz 14, I don't remember exactly.
For Fritz 15, it's the same engine as Deep Fritz 15.

2) Don't you think these 2 engines are the same one : "Junior 12" and "Junior 120" ?

Some more remarks :

1) You merged 2 versions of Movei : "0.08.438" and "0.08.438 P10", but I think the "P10" version gave a remarkable Elo boost, this 2 versions shouldn't be merged.

2) For Shredder and Junior you can merge 32bit and 64bit versions as the 64bit compile doesn't give Elo boost.

3) What's the difference between "Rybka 4" and "Rybka 4*" ?

Rybka 4* is 'Fritz' 15 (I thought I mentioned it somewhere)

Fritz 15 is cleraly not the same engine as Rybka 4, it's the same author but many year laters and stronger (around 25 Elo). It's really strange to use the same name for 2 different engines.

Code: Select all

cegt :
Fritz 15 x64 1CPU       	3018
Deep Rybka 4.1 x64 1CPU 	2979

ccrl :
Fritz 15 64-bit 	3105	
Rybka 4.1 64-bit	3085	
Rybka 4 64-bit  	3083	

I don't think P10 or whatever setting gave statistical relevant differences,
moreover it seems it was nearly always used - I just omitted it...
like 'contempt' settings.

Yes, there's a relevant difference, around 25 rating points, and the 2 big lists make a clear difference between them with hundreds of games :

Code: Select all

cegt :
Movei 0.08.438 P10 	2461 	1319
Movei 0.08.438  	   2438 	1678

ccrl :
Movei 00.8.438 (10 10 10)	2667	2639
Movei 00.8.438	           2645	1009

GURL 20160414 :
Movei 0.08.438 P10   :  2719.1     4013    
Movei 0.08.438       :  2678.9     8032

For the SSDF question did you read the comment in the introduction?
SSDF could be only included for the games when both opponents played
the same hardware, otherwise it would falsify the results.
(and I started with a minimum hardware of P90 IIRC - and no arm/board
hardware etc.)
The SSDF games were all filtered therefore before.

Yes, I did read your comment, did you read mine ?


About Fritz, I gathered some evidence you shouldn't mix "Deep" and "not-Deep" version, it's not the same code as it was release months after :

Release Dates :

Code: Select all

Fritz 7 : 2001-07
Deep Fritz 7.0 : 2003

Fritz 8 : 2003-9
Deep Fritz 8.0 : 2005

Fritz 10 : 2006-10
Deep Fritz 10 : 2006-12

Fritz 11 : 2007-11
Deep Fritz 11 : 2008-11

Fritz 12 : 2009-10
Deep Fritz 12 : 2010-7

Fritz 13 : 2011-10
Deep Fritz 13 : 2012-06

Ratings+number of games

Code: Select all

 761 Fritz 11                  2937  10857 
 668 Deep Fritz 11             2960   1747 
 
 662 Fritz 12                  2962   7714 
 562 Deep Fritz 12             2992   3283 
 
 461 Fritz 13                  3036  12167 
 445 Deep Fritz 13             3041   1296 

[Vinvin]

...
It would be nice if you or someone else could run the file with the given batch (200 simuls) and send me the resulting file.
(I will add then those numbers - currently the error numbers are only run with a sparse 10 simuls due to lack of processor time here)

I've started the .bat. How much time will it take ?

Please run it again "

I ran the batch as it is (200 simul) and it took 17 hours to complete.
I'll run after I'll be finished to search for more inaccuracies in the list, that take a bit of time

After some research I finally merged WChess 1.6 and 1.06 to 1.6 and of course Junior 12 and 120 to 12.
The updated file is uploaded already.

Nice !

(I will not merge 32/64 bit versions for some programs even if they show practically no difference - after all I separated the columns for bits/cores to give that info)

If an engine show no speed improvement from 32b to 64b, it's a non sense to duplicate it.
IIRC, it's the case for Shredder but I don't want to spend more time to find this in the archives.

For 50 simuls the time needed was once nearly 4 hours on 1 core on my 2009 quadcore.

OK, I'll run for 50 simuls next time, that should take 4 hours.
It's pity multi-thread is not used there, that could divide the time by 10.

[Guenther]

3) What's the difference between "Rybka 4" and "Rybka 4*" ?

Code: Select all

Rybka 4* is 'Fritz' 15 (I thought I mentioned it somewhere)

Fritz 15 is cleraly not the same engine as Rybka 4, it's the same author but many year laters and stronger (around 25 Elo). It's really strange to use the same name for 2 different engines.

Code: Select all

cegt :
Fritz 15 x64 1CPU       	3018
Deep Rybka 4.1 x64 1CPU 	2979

ccrl :
Fritz 15 64-bit 	3105	
Rybka 4.1 64-bit	3085	
Rybka 4 64-bit  	3083	

I wrote 'Rybka 4*' (not 'Rybka 4') and there is a lot of evidence that it is a (not much further) developed version of Rybka 4.1 and it has also all the same options... the strength increase also is very modest and moreover they finally had given up to use that fossile CB protocol, which
died like a last dinosaur and was long expected. So they could just take Rybka as it was and stick their name on it.

I also don't like to name Fritz 14 to Fritz as it was done by Gyula Horvath (incorporating a private version of a program which never could be inspected before) and there was even evidence in sim tests that it should be marked doubtful and has nothing in common with previous Fritz versions.

For Movei I can say, or did already (?), that this P10 setting was already long known before that version (Uri always wanted to have that one used) and AFAIK most people already used it WITHOUT adding another pompous suffix to it thus both entities were already mixed before... (I cannot separate them again anyways...)

Maybe I just keep the list private again - I have not much intention to repeat or add more boring work into it... I can also rename it if you don't like it... probably more harm was already done by wrong naming of programs in the original sources...

I have a lot of arguments for what I did but I am bored to write them all down and a lot of them are originated in heavy knowledge of all what happened with versions and programs from past 'til today and I am not in the mood for lengthy explanations for things that happened long before this forum in that form even existed.

[Guenther]

OK, I'll run for 50 simuls next time, that should take 4 hours.
It's pity multi-thread is not used there, that could divide the time by 10.

What do you mean by 'there'? Ordo can be run with mutiple cores, but
I just needed the other cores for other things not related to chess.
(Or do you mean you have only one core?)

[Vinvin]

OK, I'll run for 50 simuls next time, that should take 4 hours.
It's pity multi-thread is not used there, that could divide the time by 10.

What do you mean by 'there'? Ordo can be run with mutiple cores, but
I just needed the other cores for other things not related to chess.
(Or do you mean you have only one core?)

I mean Ordo uses only 1 core, it's pity multi-thread is not used there, that could divide the time by 10.

[Guenther]

OK, I'll run for 50 simuls next time, that should take 4 hours.
It's pity multi-thread is not used there, that could divide the time by 10.

What do you mean by 'there'? Ordo can be run with mutiple cores, but
I just needed the other cores for other things not related to chess.
(Or do you mean you have only one core?)

I mean Ordo uses only 1 core, it's pity multi-thread is not used there, that could divide the time by 10.

oops you can! there is a flag for multi core usage during calculation of the simuls. -n 'number of cores' (this is from the pdf manual, but I found no prog in hurry which extracts text from it, otherwise I would give you the exact quote)

[Vinvin]

OK, I'll run for 50 simuls next time, that should take 4 hours.
It's pity multi-thread is not used there, that could divide the time by 10.

What do you mean by 'there'? Ordo can be run with mutiple cores, but
I just needed the other cores for other things not related to chess.
(Or do you mean you have only one core?)

I mean Ordo uses only 1 core, it's pity multi-thread is not used there, that could divide the time by 10.

oops you can! there is a flag for multi core usage during calculation of the simuls. -n 'number of cores' (this is from the pdf manual, but I found no prog in hurry which extracts text from it, otherwise I would give you the exact quote)

Yes, I found in the doc here : https://docs.google.com/viewer?a=v&pid= ... OGE2NDRlMA
But the multi-thread is not very efficient on my 6 cores, I set "-n 6" but ordo only use 8% to 16% CPU time, it seems blocked at 2 threads

[Guenther]

Yes, I found in the doc here : https://docs.google.com/viewer?a=v&pid= ... OGE2NDRlMA
But the multi-thread is not very efficient on my 6 cores, I set "-n 6" but ordo only use 8% to 16% CPU time, it seems blocked at 2 threads :-(

Out of curiosity I started a quick run with '-n 3' and it seems the cpu usage changed later (to 50% - only checked for a few minutes) and the memory usage jumped by factor 3 or 4 at the beginning of the simuls already.

OTH there is a disadvantage in displaying progress now. Before the current number of simuls could be seen, now I only see a progress chart which doesn't change (at least while I was watching).

[Vinvin]

...
It would be nice if you or someone else could run the file with the given batch (200 simuls) and send me the resulting file.
(I will add then those numbers - currently the error numbers are only run with a sparse 10 simuls due to lack of processor time here)

I've started the .bat. How much time will it take ?

Please run it again

After some research I finally merged WChess 1.6 and 1.06 to 1.6 and of course Junior 12 and 120 to 12.
The updated file is uploaded already.

(I will not merge 32/64 bit versions for some programs even if they show practically no difference - after all I separated the columns for bits/cores
to give that info)

For 50 simuls the time needed was once nearly 4 hours on 1 core on my 2009 quadcore.

There still some issues with lowercase/uppercase with :

Code: Select all

discocheck
grapefruit
greko
nagaskaki
obender
smarthink
wildcat

Example :

Code: Select all

"DiscoCheck"	"403"	"x64"		2756,4
"Discocheck"	"403"	"x64"		2715,8

[Guenther]

There still some issues with lowercase/uppercase with :

Code: Select all

discocheck
grapefruit
greko
nagaskaki
obender
smarthink
wildcat

Example :

Code: Select all

"DiscoCheck"	"403"	"x64"		2756,4
"Discocheck"	"403"	"x64"		2715,8

Thanks! I will repair it, but probably not today, because I will leave the town again soon.