New LCT II results
Frederic LOUGUET
Programs LCT II POS TAC FIN Hash
rating % % %
Fritz 4 2455 43 69 46 16 MB
Genius 4 2470 35 68 67 16 MB
Hiarcs 4 2505 57 61 54 64 MB
M Chess 5 2440 42 61 54 10 MB
Rebel 7 2510 56 67 50 60 MB
Virtual Chess 2430 51 65 30 16 MB
Crafty 9.21 2340 31 49 50 832 KB
The complete LCT II test with all detailed results, explanations and positions in a lot of
formats (Genius, Fritz, Hiarcs, Rebel, Virtual Chess, Socrates, Kallisto, Crafty,
Chessmaster 4000, Gideon, M Chess Pro, W Chess and Zarkov) is available at Mads
Brevik Chess Page :
http://knoll.kih.no:8001/stud/mads2/chess.htm.
I never tested GNU Chess with the LCT II, so I would appreciate if someone could do it
and send me the results.
The current version of the test is 1.20, with results on Pentium Pro. I can also send you the
tests in your mail, just drop me a word about it at : lou...@worldnet.net.
--
Frederic Louguet
lou...@worldnet.net
Frederic LOUGUET
DX4 Notebook, no L2 cache, 10 MB hash : 2130 (6333 nps on first test - LCTPOS01 d6!)
P90, 256 KB asynchronous L2 cache, 7 MB hash : 2235 (13048 nps)
P133, 256 KB pipeline burst L2 cache, 14 MB hash : 2305 (24505 nps)
PP150, 16+256 KB internal caches, 56 MB hash : 2365 (40826 nps)
PP200, 16+256 KB internal caches, 56 MB hash : 2395 (54888 nps)
On Pentium Pro 200 (PP200) :
-------------------------------------------
Rebel 7 : 2510 (60 MB hash)
Hiarcs 4 : 2505 (64 MB hash)
Genius 4 : 2470 (16 MB hash)
Fritz 4 : 2455 (16 MB hash)
M Chess Pro 5 : 2440 (10 MB hash)
Virtual Chess : 2430 (16 MB hash)
It would be interesting to know results of Crafty on Alphas, and on Cray !
--
Frederic Louguet
lou...@worldnet.net
Robert Hyatt
Frederic LOUGUET <lou...@worldnet.net> wrote:
-->Some results on Crafty 9.21 and the LCT II chess test
-->
-->DX4 Notebook, no L2 cache, 10 MB hash : 2130 (6333 nps on first test - LCTPOS01 d6!)
-->P90, 256 KB asynchronous L2 cache, 7 MB hash : 2235 (13048 nps)
-->P133, 256 KB pipeline burst L2 cache, 14 MB hash : 2305 (24505 nps)
-->PP150, 16+256 KB internal caches, 56 MB hash : 2365 (40826 nps)
-->PP200, 16+256 KB internal caches, 56 MB hash : 2395 (54888 nps)
-->
-->On Pentium Pro 200 (PP200) :
-->-------------------------------------------
-->Rebel 7 : 2510 (60 MB hash)
-->Hiarcs 4 : 2505 (64 MB hash)
-->Genius 4 : 2470 (16 MB hash)
-->Fritz 4 : 2455 (16 MB hash)
-->M Chess Pro 5 : 2440 (10 MB hash)
-->Virtual Chess : 2430 (16 MB hash)
-->
-->It would be interesting to know results of Crafty on Alphas, and on Cray !
-->
The real problem issue here (I think) is less of speed, and more about
evaluation. Crafty can improve in this respect quite a bit, although I'm
not sure that it will over the short-haul. I'm still working on the "anti-
computer" strategies, and this frequently is at odds with what a GM would say
is *the* best move. The interesting challenge will be to "smooth" the eval
so that the anti-computer strategies "kick in" when the opponent tries to
do one of the usual anti-computer things, while not having them affect the
move choice for normal-type positions. Seems to be doable, but the line is
*very* fine here...
I've watch Crafty in many games now, and, while it's a long-way from being
"perfect" it has begun to avoid a lot of the "classic mistakes" like locking
pawns. It is still weaker than I'd like in analyzing king safety, and also
(like every computer program I've seen) has a propensity to attack on the
queen-side while the king-side burns down around its ears. It's solvable,
but I've watch Crafty get worse at (say) the CCR rating suite, while playing
significantly better against "computer buster" type players. The key is can
this be "balanced" better? Time will tell...
Bob
--
Robert Hyatt Computer and Information Sciences
hy...@cis.uab.edu University of Alabama at Birmingham
(205) 934-2213 115A Campbell Hall, UAB Station
(205) 934-5473 FAX Birmingham, AL 35294-1170