Great endgame test for dedicated machines (hash table)

[JonP01]

I came across this interesting position in one of my machine versus machine games. Once this position was reached, I waited for what I thought would be the obvious winning move - Rc2 (there are clearly a few other "correct" moves as well where the King covers b1, but Rc2 is the one I would have played).

I was very surprised when white infact played Rxe6?? This move pretty much converts a completely won position to an extremely difficult win at best and perhaps nothing more than a draw (the game was infact drawn). When, after the game I put the position through HIARCS to verify I was not missing something, the evaluation went instantly from 5.83 in favour of white to only 1.43 in favour of white after Rxe6??

I then tried this position on my most powerful dedicated (Novag Amber) and it too plays Rxe6 even on it's longest tournament level. And even on analysis, it still wasn't budging from Rxe6.

I then tried the Mephisto Roma emulation on my Palm (which runs 3 times the speed of the original Roma) and it too could not find anything better than Rxe6 - even after 6 minutes thought. For those who don't know, Roma was the last top of the line Mephisto that did not have hash tables.

Contrast this to Chess Genius on the same Palm machine (with 200 Mhz xscale processor) and it found one of several correct continuations after only 25 seconds. But of course Chess Genius has a very fast processor and hash tables - 2Mb worth of them. It would have been interesting to see what it thought without hash tables, but they can't be disabled.

Anyway, I am curious to see if there is any dedicated chess machine in existence that does not have endgame hash tables and can find one of the correct continuations in this position within a reasonable time frame (by that I mean the maximum time it might take in a tournament game - perhaps ten minutes or less)? I wonder if such a machine exists? And of course, I would be interested to know how long it takes models with hash tables as well. I wonder, for example, can machines like Novag Diamond / Sapphire or StarDiamond find this on their 40 move in 2 hour tournament levels?


[D]5k2/8/1PR1p1p1/3p3p/8/1K4P1/7P/5r2 w - - 0 50

[Steve B]

[D]5k2/8/1PR1p1p1/3p3p/8/1K4P1/7P/5r2 w - - 0 50

Anyway, I am curious to see if there is any dedicated chess machine in existence that does not have endgame hash tables and can find one of the correct continuations in this position within a reasonable time frame (by that I mean the maximum time it might take in a tournament game - perhaps ten minutes or less)? I wonder if such a machine exists?

Hi Jon
as i just happened to have the Fidelity Elegance's out for testing and their annual Physicals i loaded up your position

i can report that both the Elegance A(3Mhz) and the Elegance B(3.6 Mhz)play Rc2 after 18 minutes and 15 minutes respectively

And of course, I would be interested to know how long it takes models with hash tables as well. I wonder, for example, can machines like Novag Diamond / Sapphire or StarDiamond find this on their 40 move in 2 hour tournament levels?

on tournament level 40/2...the Sapphire(I) plays Rc2 returning a PV of +4.07 and reaching 8 ply

i am curious to see how accurate your Roma Emulation is so if i get some time i will load the position up on the Roma 16 modules
Big Game Tonight Regards
PS.The Brothers Parle have me on my toes today
just finished some work for Cameron on the Yeno 430XT he is interested in

[Steve B]

[D]5k2/8/1PR1p1p1/3p3p/8/1K4P1/7P/5r2 w - - 0 50

i am curious to see how accurate your Roma Emulation is so if i get some time i will load the position up on the Roma 16 modules

ok guess i found the time


i can verify that BOTH the Roma 16 AND 32 modules play Rxe6 after 6 minutes and even after 15 minutes

proof positive once again that the Spracklens were far far better programmers then Lang ever was

Regards
Steve
PS.Speaking of Tonight's big showdown..i fed the Tasc R30( v2.2) the position as well(which by now i know by heart)
i set it to NORMAL which is better for positions arising from actual games rather then ACTIVE(which is better for solving "Problem like "positions)

R30 picks Rc2 after 106 seconds -7 ply and searching just under 515000 positions

[JonP01]

Hi Steve,

Thanks for the feedback. I suspected a Fidelity machine might get it right, but alas it looks likely it isn't going to do it on the normal tournament level, since the thinking time was likely too long. But the fact that they get it at all is pretty amazing. I wonder what the Excel 68000 does it in? Maybe Cameron can answer that one. I actually left the Roma emulation going for well over an hour and it still did not find the right move.

Good to see the top of the line Novags can find it though - one would have hoped they could, but that is still a good effort. Given the much slower hardware on the Sapphire compared to my Palm in ARM mode, finding the correct move in tournament time versus 25 seonds for the Palm is a very impressive.

I might just see what happens if I give my Chessmaster GBA about 20 minutes to think about this, since it was the one playing white in this particular game.


Jonathan

[Terry McCracken]

Hi Steve,

Thanks for the feedback. I suspected a Fidelity machine might get it right, but alas it looks likely it isn't going to do it on the normal tournament level, since the thinking time was likely too long. But the fact that they get it at all is pretty amazing. I wonder what the Excel 68000 does it in? Maybe Cameron can answer that one. I actually left the Roma emulation going for well over an hour and it still did not find the right move.

Good to see the top of the line Novags can find it though - one would have hoped they could, but that is still a good effort. Given the much slower hardware on the Sapphire compared to my Palm in ARM mode, finding the correct move in tournament time versus 25 seonds for the Palm is a very impressive.

I might just see what happens if I give my Chessmaster GBA about 20 minutes to think about this, since it was the one playing white in this particular game.


Jonathan

Spracklen's programs at the time were ahead of Lang and Kittenger in the endings.

Kittenger wasn't know for endgame ability in the 80's or early 90's.

Lang too lagged a bit in this area, but the middlegame play was somewhat better than Spracklen's programs, while Kittenger's programs at the time were the most human like in the middlegame was trailing a bit behind in results. Nevertheless, his machines appeared to give the most interesting play in the middlegame and in cases of pawn storms and King-side attacks his programs were for their time outstanding.

Spracklen's programs didn't know what a pawn storm was!

But was very strong in piece play, most of the time by comparison.

Terry