Schachcomputer.info Community

Zurück   Schachcomputer.info Community > Schachcomputer / Chess Computer: > Die ganze Welt der Schachcomputer / World of chess computers


Antwort
 
Themen-Optionen Ansicht

  #1  
Alt 23.02.2014, 09:17
Benutzerbild von IvenGO
IvenGO IvenGO ist offline
Fidelity Elite Avantgarde Version 10
 
Registriert seit: 19.05.2012
Ort: Moscow
Land:
Beiträge: 213
Abgegebene Danke: 241
Erhielt 235 Danke für 101 Beiträge
Aktivitäten Langlebigkeit
0/20 12/20
Heute Beiträge
0/3 ssssss213
Beitrag Hash for DCCs

Hi all,

at first some thoughts about influense of subject for DCCs playing strength that I have:

Atlanta (+512kb) - Milano Pro
2267 - 2160 (german ELO)
2274 - 2154 (dutch ELO)

...so here is ~+100 ELO if +512kb Hash Tables.

Next I took my London 16-bit and turn Hash=OFF and got these match results (30min/game):

draw 5:5 vs Scorpio
lost 4:6 vs FD2265
won 5.5:4.5 vs Milano

seems like strength could be claimed around 2050, so again +512kb hash gives ~+100 ELO gain in active chess, this time for other programmer's computer!

And there are the questions:

Which amount of RAM was rated generally for RISC 2500?
Which ELO difference might be predicted between 128kb and 512kb/2mB versions?

Thanks in advance!
Mit Zitat antworten
  #2  
Alt 23.02.2014, 12:16
Benutzerbild von Chessguru
Chessguru Chessguru ist offline
Administrator
 
Registriert seit: 29.06.2004
Ort: Rostock
Alter: 55
Land:
Beiträge: 6.479
Bilder: 1
Abgegebene Danke: 1.589
Erhielt 4.600 Danke für 1.381 Beiträge
Member Photo Albums
Aktivitäten Langlebigkeit
10/20 20/20
Heute Beiträge
2/3 sssss6479
AW: Hash for DCCs

Hi,

 Zitat von IvenGO Beitrag anzeigen
And there are the questions:

Which amount of RAM was rated generally for RISC 2500?
Which ELO difference might be predicted between 128kb and 512kb/2mB versions?
on active chess, no relevant differences...


A look at the CSS (Computerschach und Spiele / Heft 1/94 / S.13-16):



and




The diagramms:

  • 1...Kb8 - remis



  • 1.Nb6 - white wins



  • 1.Kg3 - remis



  • 1.Kh8 - remis



  • 1.Kb1 - white wins



  • 1.c7 - white wins



  • 1.Ke6 - remis



  • 1.Kg6 - remis



  • 1.Ba7 - white wins



BT-2630-Test / Schachcomputer.info Wiki

Risc 2500 BT2630 Test / 2 MB and 512 KB




Best regards,
Micha
Mit Zitat antworten
Folgende 3 Benutzer sagen Danke zu Chessguru für den nützlichen Beitrag:
Boris (23.02.2014), Eckehard Kopp (23.02.2014), IvenGO (24.02.2014)
  #3  
Alt 23.02.2014, 16:08
Benutzerbild von EberlW
EberlW EberlW ist offline
Lebende Foren Legende
 
Registriert seit: 09.01.2005
Ort: Leverkusen-Küppersteg
Alter: 58
Land:
Beiträge: 3.111
Abgegebene Danke: 45
Erhielt 58 Danke für 43 Beiträge
Aktivitäten Langlebigkeit
0/20 20/20
Heute Beiträge
0/3 sssss3111
AW: Hash for DCCs

 Zitat von Chessguru Beitrag anzeigen
on active chess, no relevant differences...
Fully agreed!

A few years ago I tested several engine versions of "King" and "Gideon" on different chess machines (128k, 512k and 1024k) and found no reasonable differences in solving times. No matter if the tested positions were midgames or endings. I did not try these positions with hash disabled but of course the solving times would noticeable increase in this case. It seems, the amount of hash doesn't matters much for most engines, as long as it is at least 128k on DCC's. Even if one can find the one or the other position where different hash sizes make a big difference in solving times, this will not have any noticeable influence to the overall playing strength of DCC's.

But of course this is only my personal opinion and i won't claim it the one and only truth in concerns to DCC's.
Kind regards, Wilfried
Mit Zitat antworten
Folgender Benutzer sagt Danke zu EberlW für den nützlichen Beitrag:
IvenGO (24.02.2014)
  #4  
Alt 24.02.2014, 02:46
Benutzerbild von IvenGO
IvenGO IvenGO ist offline
Fidelity Elite Avantgarde Version 10
 
Registriert seit: 19.05.2012
Ort: Moscow
Land:
Beiträge: 213
Abgegebene Danke: 241
Erhielt 235 Danke für 101 Beiträge
Aktivitäten Langlebigkeit
0/20 12/20
Heute Beiträge
0/3 ssssss213
Re: Hash for DCCs

Hmm... but test positions are taken from human games and the practice of DCC vs DCC is a bit different thing...

Some more results:

Star Diamond (Hash=ON) - Master Chess 7 : 3
Star Diamond (Hash=OFF) - Master Chess 6 : 4
Star Diamond (Hash=ON) - Atlanta 2.5 : 7.5

So we can see here that 118kb didnt have big effect but 512kb completely changed the situation on chess board!

I forgot to say that in all matches there were the same opening lines played for more accurate results.
Mit Zitat antworten
Folgender Benutzer sagt Danke zu IvenGO für den nützlichen Beitrag:
Theo (24.02.2014)
  #5  
Alt 24.02.2014, 07:06
Benutzerbild von EberlW
EberlW EberlW ist offline
Lebende Foren Legende
 
Registriert seit: 09.01.2005
Ort: Leverkusen-Küppersteg
Alter: 58
Land:
Beiträge: 3.111
Abgegebene Danke: 45
Erhielt 58 Danke für 43 Beiträge
Aktivitäten Langlebigkeit
0/20 20/20
Heute Beiträge
0/3 sssss3111
AW: Re: Hash for DCCs

 Zitat von IvenGO Beitrag anzeigen
Hmm... but test positions are taken from human games and the practice of DCC vs DCC is a bit different thing...
Of course there is a difference between human test positions and practical computer games. Nonetheless hash tables will show effect in both cases. How much effect only depend on the actual given position and even more on the way HOW the programmer told his engine to make use of this tables. I am no programmer and can't explain how the one or the other system works, but i guess you will agree that nowadays PC engines use their hash tables in a different way than DCC's. There are differences inbetween DCC's, too.
Zitieren:
Some more results:

Star Diamond (Hash=ON) - Master Chess 7 : 3
Star Diamond (Hash=OFF) - Master Chess 6 : 4
Star Diamond (Hash=ON) - Atlanta 2.5 : 7.5

So we can see here that 118kb didnt have big effect but 512kb completely changed the situation on chess board!

I forgot to say that in all matches there were the same opening lines played for more accurate results.
Oh, i saw the comparison between Atlanta and Milano Pro before, but now i think it is time to mention, that there are more differences between Atlanta and Milano Pro (Master Chess) than only the hash tables. At first, the Atlanta's hardware is noticeable faster than the others. At second, the Atlanta has a later engine version. If you compare Atlanta to Milano Pro (Master Chess) it is something similar like comparing Fritz 3 / P90 vs. Fritz 2 / 486-100. Of course the Atlanta don't has Fritz 3 as engine - i just wanted to explain in an example, what you are trying to compare.

Kind regards, Wilfried
Mit Zitat antworten
Folgender Benutzer sagt Danke zu EberlW für den nützlichen Beitrag:
Theo (24.02.2014)
  #6  
Alt 24.02.2014, 08:06
Benutzerbild von IvenGO
IvenGO IvenGO ist offline
Fidelity Elite Avantgarde Version 10
 
Registriert seit: 19.05.2012
Ort: Moscow
Land:
Beiträge: 213
Abgegebene Danke: 241
Erhielt 235 Danke für 101 Beiträge
Aktivitäten Langlebigkeit
0/20 12/20
Heute Beiträge
0/3 ssssss213
Re: Hash for DCCs

Zitieren:
At first, the Atlanta's hardware is noticeable faster than the others.
Ehh... what do You mean? Schach-computer.info/wiki says all 3 models have the same hardware (SH7034 at 20Mhz) except hash tables bonus for Atlanta. Is it incorrect information?

And about Fritz-3 inside: an owner of MilanoPro told me that on the box there's a writting kind of "contains WCCC-1995 champion programm"
Mit Zitat antworten
  #7  
Alt 24.02.2014, 09:16
Benutzerbild von Robert
Robert Robert ist offline
Lebende Foren Legende
 
Registriert seit: 30.06.2004
Ort: Regensburg
Alter: 60
Land:
Beiträge: 4.179
Abgegebene Danke: 1.782
Erhielt 783 Danke für 464 Beiträge
Aktivitäten Langlebigkeit
6/20 20/20
Heute Beiträge
0/3 sssss4179
AW: Re: Hash for DCCs

Hi Willi and IvenGO,

 Zitat von EberlW Beitrag anzeigen
Of course there is a difference between human test positions and practical computer games.
Hm, i just wanted to clear up something:
If you set up a position, the hash tables have to be filled from scratch, which costs some time, while in a practical game, they are already filled, when you reach that position.
Is this, what you both meant?

Would be really interesting in my eyes, if there are noticable differences, when the BT-positions would be reached in practical games...


Regards,
Robert
Mit Zitat antworten
  #8  
Alt 24.02.2014, 19:29
Benutzerbild von EberlW
EberlW EberlW ist offline
Lebende Foren Legende
 
Registriert seit: 09.01.2005
Ort: Leverkusen-Küppersteg
Alter: 58
Land:
Beiträge: 3.111
Abgegebene Danke: 45
Erhielt 58 Danke für 43 Beiträge
Aktivitäten Langlebigkeit
0/20 20/20
Heute Beiträge
0/3 sssss3111
AW: Re: Hash for DCCs

 Zitat von IvenGO Beitrag anzeigen
Ehh... what do You mean? Schach-computer.info/wiki says all 3 models have the same hardware (SH7034 at 20Mhz) except hash tables bonus for Atlanta. Is it incorrect information?
I see (haven't had a look to this base for a while) and still disagree. If we look at the SH7034 entry of our wiki, it shows several models - not only those w-w/o "hash". A lot depends on the "suffix", means the small term behind "SH7034". I really have to admit, that i dindn't had a look into my Atlanta or Milano Pro for a very long time and can't remember what type of CPU's are built in there. I will try my best to open both computers cases during next Saturday to find out the "true" CPU types, but i can't promise anything. Meanwhile i'd like you to answer a few small questions for yourselfes:

If it is possible to speed up an Atlanta from 20 to 40 MHz, why it is impossible to do the same with a Milano Pro or Master Chess? Both will end at approximately 27 MHz.
If you speed up an Atlanta and a Milano Pro (Master Chess), why is the Atlanta's clock running faster while the Milano Pro's clock still runs at real time?
If one or even both of the questions above get answered with "I dunno!", why should we assume these CPU's to be equal?

Please let us delay this discussion for a couple of days. On weekend i will try to find an answer to the CPU issue - during the week i'm unfortunately much to tired...
Zitieren:
And about Fritz-3 inside: an owner of MilanoPro told me that on the box there's a writting kind of "contains WCCC-1995 champion programm"
Well, the box and the stickers can tell a lot...
Guido (ed_209) and i tried almost everything to find a prove for the engine versions of Atlanta and Milano Pro by comparison to any available Fritz version from 1.0 to 3.3 (or was it just 3.08?) and we found no final solution, even if we tried computers from 386SX-16 up to modern PC's - the ladder with several slow motions of course. Fritz 3 is noticeable stronger than Atlanta and by far stronger than Milano Pro - these engines are different for sure! A very close comparison but still no real "hit" is:

Milano Pro = a very early Fritz 2
Atlanta = something between Fritz 2.51 and Fritz 3 (but closer to the earlier version)

Maybe our problem to find a "match" is because the PC's have different access times to their memory and different times to work instructions. Furthermore we don't know how the engines were ported from system A to system B. If Frans used different compilers and languages (both most probably), we will never find any comparison. That's why we finally gave up on searching.

Kind regards, Wilfried
Mit Zitat antworten
  #9  
Alt 24.02.2014, 19:44
Benutzerbild von EberlW
EberlW EberlW ist offline
Lebende Foren Legende
 
Registriert seit: 09.01.2005
Ort: Leverkusen-Küppersteg
Alter: 58
Land:
Beiträge: 3.111
Abgegebene Danke: 45
Erhielt 58 Danke für 43 Beiträge
Aktivitäten Langlebigkeit
0/20 20/20
Heute Beiträge
0/3 sssss3111
AW: Re: Hash for DCCs

 Zitat von Robert Beitrag anzeigen
Hi Willi and IvenGO,


Hm, i just wanted to clear up something:
If you set up a position, the hash tables have to be filled from scratch, which costs some time, while in a practical game, they are already filled, when you reach that position.
Is this, what you both meant?

Would be really interesting in my eyes, if there are noticable differences, when the BT-positions would be reached in practical games...


Regards,
Robert
Hi Robert,

as it concerns me, i meant this almost like you explained.
The only difference: Some early hash algo's cleared the RAM after every move and in this case there is no difference to position testing.

To your second statement: I'm not sure weather the one or the other BT-position hasn't come from a practical game?

Will you participate in Klingenberg this year? As for me, i will try to get a few days off and to join this event. Maybe we will find a few minutes to discuss this face by face in our native language (although i'm not that sure that anyone else will understand what we are talking about, after we had some quarters of red wine... )

Drunken Chess Master regards, Wilfried
Mit Zitat antworten
  #10  
Alt 25.02.2014, 09:04
Benutzerbild von Robert
Robert Robert ist offline
Lebende Foren Legende
 
Registriert seit: 30.06.2004
Ort: Regensburg
Alter: 60
Land:
Beiträge: 4.179
Abgegebene Danke: 1.782
Erhielt 783 Danke für 464 Beiträge
Aktivitäten Langlebigkeit
6/20 20/20
Heute Beiträge
0/3 sssss4179
AW: Re: Hash for DCCs

 Zitat von EberlW Beitrag anzeigen
as it concerns me, i meant this almost like you explained.
The only difference: Some early hash algo's cleared the RAM after every move and in this case there is no difference to position testing.
Ah, i see! But do we know, which computers are clearing their RAM? I'm afraid, we don't...
Zitieren:
To your second statement: I'm not sure weather the one or the other BT-position hasn't come from a practical game?
I remember one special case in an Online-WM: Steffen had played with his Tiger Grenadier and around move 70 it made a terrible mistake. After the game he set up the position in his Grenadier and then it played a different (better) move! He asked me to verify this and my Grenadier also played this move when i set up the position. But then i replayed the game from the start and now my Grenadier blundered just like Steffens did in the actual game!

Even without hashtables it sometimes seems to make a difference; and with hash this should happen much more often...
Zitieren:
Will you participate in Klingenberg this year? As for me, i will try to get a few days off and to join this event. Maybe we will find a few minutes to discuss this face by face in our native language (although i'm not that sure that anyone else will understand what we are talking about, after we had some quarters of red wine... )
Nee, leider nicht; ich habe einen Termin, den ich schlecht sausen lassen kann.

Echt schade, denn nachdem ich letztes Jahr krankheitsbedingt schon nicht konnte, wäre ich dieses Jahr gern dabei gewesen! Vor allem, weil ein paar Leute dabei sind, die ich bislang nur aus dem Forum kenne. Und wir beide haben uns ja auch schon lange nicht mehr gesehen!

Naja, vielleicht ein anderes Mal...


viele Grüße
Robert
Mit Zitat antworten
Antwort

Themen-Optionen
Ansicht

Forumregeln
Du bist nicht berechtigt, neue Themen zu erstellen.
Du bist nicht berechtigt, auf Beiträge zu antworten.
Du bist nicht berechtigt, Anhänge hochzuladen.
Du bist nicht berechtigt, deine Beiträge zu bearbeiten.

BB code ist An
Smileys sind An.
[IMG] Code ist An.
HTML-Code ist An.

Gehe zu

Ähnliche Themen
Thema Erstellt von Forum Antworten Letzter Beitrag
Review: WMCCC 6./7./8. Elozahlen, Hash und vieles mehr José Die ganze Welt der Schachcomputer / World of chess computers 9 07.09.2011 11:31


Alle Zeitangaben in WEZ +1. Es ist jetzt 16:42 Uhr.



Powered by vBulletin (Deutsch)
Copyright ©2000 - 2024, Jelsoft Enterprises Ltd.
©Schachcomputer.info