gluarung beta versus bright .3a

[ozziejoe]

4 min 2 sec. dual core, all engines using two threads. noomen 30 opening test set


Glaurung beta looks like a big improvement here, over epsilon 5

glaurungs versus bright-1 2008

bright-0.3a - Glaurung 2-epsilon/5 38.5 - 21.5 +27/-10/=23 64.17%
bright-0.3a - Glaurung 080420 26.5 - 33.5 +19/-26/=15 44.17%


Here is how Glaurung beta has performed overall in the tourni, compared to epsilon 5


1 Rybka 2.3.2a mp 32-bit : 3022 60 58 120 79.6 % 2786 27.5 %
2 Toga II 1.4 beta5c : 2900 53 52 120 65.8 % 2786 33.3 %
3 Glaurung 080420 : 2839 43 44 180 40.0 % 2910 28.9 %
4 bright-0.3a : 2815 52 52 120 54.2 % 2786 31.7 %
5 Glaurung 2-epsilon/5 : 2737 43 44 180 26.9 % 2910 32.8 %


So glaurung is "only" about 60 pnts off of toga as the number 2 (though this estimate is not terribly reliable till i get a lot more games)

A nice improvement so far

Can someone tell me (tord?) how this glaurung beta compares to the glaurung 2.1 ccrl is testing?

best
J

[swami]

Glaurung latest beta in my estimation is over 70 elo improvement in blitz alone.

[Eelco de Groot]

Hello Joseph,

Thanks very much for these results! That is looking very good for the new Glaurung! Maybe Tord can say something about it but maybe he is busy with putting the final touches on Glaurung, probably that will be something like Glaurung 2.1? The version that you are referring to as being tested by Graham and the CCRL is I think Glaurung 2.0.1. For this version Apple MacOSX and Linux bimnaries were downloadable from Tord's site already since last december, but the best Windows versions to my knowledge are a bit harder to find, a link to some of those was posted on the forum here. I refer to the thread in this old post on the Rybka forum:

http://rybkaforum.net/cgi-bin/rybkaforu ... l=Glaurung

Hello,

There was a problem with a too long uci_engineabout string. Tord has not yet put up an official version, there are still problems with the Mac compiles I think. But the compiles by Jim Ablett should run now in Chessbase. See also:

http://www.talkchess.com/forum/viewtopi ... 23&t=17979

Rybka is a bit of an extra difficult 'Angstgegner' for Glaurung, I think mainly because Tord can't test against it on his Macs! Somebody make a Rybka for Mac version for Tord! Also Glaurung's attacking style is a bit of a risk especially when facing Rybka untested...

My latest Mjolnir settings I hope are a bit better because they are also better against Belka. (Sorry for mentioning this cloned engine here again but it was used for my first tests of the 'Mjolnir' settings and against the default Glaurung 2.0.1, so I had a lot of data against it) No wonders of course, they are just settings.

I sure would like to do some experiments with implementing Alessandro Scotti's accumulated data on material imbalances for Glaurung but my programming skills I'm afraid are not up to the task. But that would be something, Glaurung with that new knowledge and that certainly would help against real Rybka 2.3.2a too Maybe Tord has some plans in this direction as he has said before he wanted to do something like this. But getting the data implemented right does need careful adjustments I think and lots of games to test it.

Eelco

Maybe the bit about Glaurung not tested against Rybka is not very accurate anymore, and there is of course now the poor substitute possible of testing against Strelka which is only about as strong as Rybka 1.0, that is still pretty strong though and Tord can compile Osipov's cloned sources for Macs.

Note those tests I made with Mjolnir setting and to which I refer in above post from december were flawed because I had not checked the chosen openings well and Harry's .bkt book for Shredder was altered such that it kept playing much the same openings against itself Those tests were inconclusive therefore, I did not really repeat it with better settings for Harry's book or with another opening book.

I think that Glaurung 2.0.1 falls somewhere between Glaurung 2 - ε/5 and the new Glaurung 080420/080422. But 2.0.1 is certainly stronger than Glaurung 2- ε/5, certainly on new 64-bit processors in 64-bit Windows or other 64-bit operating systems, on my Athlon on Windows XP 32-bit I suspect the performance in the middlegame is slightly less, because Glaurung can't reach quite the same depths as the 64-bits version, in the endgame I really loved that version and the reached depths are already good in the 32-bits version also.

Regards, Eelco