A K+5P vs. K+6P problem
Emmanuel Marin
I am curious to know whether chess programs can find the only
correct move for White in this position. It is a study by Walkevitz,
1977 (published this week in the French newspaper Liberation).
This is a mate in 16, and the reason of the correct move appears
only at move 10.
This is the kind of problem an human can easily find (ie : I found
it :-) ), I'd like to know if it is the kind of problem computers
don't like...
White to play and win :
White : Kc7, Pa5, c2, d5, e5, g5
Black : Ka8, Pa6, b3, c4, e6, f7, h3
Emmanuel Marin
Paris, France
Howard Exner
Emmanuel Marin <Emmanue...@wanadoo.fr> wrote in article
<52nq4j$q...@cyan.wanadoo.fr>...
> I am curious to know whether chess programs can find the only
> correct move for White in this position. ...SNIP...
>
> White to play and win :
>
> White : Kc7, Pa5, c2, d5, e5, g5
> Black : Ka8, Pa6, b3, c4, e6, f7, h3
I tried this out on my three chess programs on a 486-133.
CM4000 and MCPro 3.5 were unable to solve this by the 20
minute mark so I stopped the search. I didn't expect Rebel 8.0
to find it either but to my surprise it announced mate at 1:59 !
Sacrificing the pawn by g5-g6 allows an eventual check on f8
which permits the ensuing mate. Nice problem.
Robert Hyatt
Howard Exner (hex...@dlcwest.com) wrote:
:
:
: Emmanuel Marin <Emmanue...@wanadoo.fr> wrote in article
I ran it on crafty 10.15, and at about 10 minutes it went from a
draw by repetition to g6 with a fail high. I waited another 10
minutes but got no PV, got disgusted and quit. :) However, it
would likely play g6 given 10 minutes since it failed high right
around there. With a larger hash it might be quicker, I didn't
try anything big...
I'm assuming g6 is the correct move?
Bob
Ed Schröder
"Howard Exner" <hex...@dlcwest.com> wrote:
>
>
>Emmanuel Marin <Emmanue...@wanadoo.fr> wrote in article
><52nq4j$q...@cyan.wanadoo.fr>...
>> I am curious to know whether chess programs can find the only
>> correct move for White in this position. ...SNIP...
>>
>> White to play and win :
>>
>> White : Kc7, Pa5, c2, d5, e5, g5
>> Black : Ka8, Pa6, b3, c4, e6, f7, h3
>
>I tried this out on my three chess programs on a 486-133.
>CM4000 and MCPro 3.5 were unable to solve this by the 20
>minute mark so I stopped the search. I didn't expect Rebel 8.0
>to find it either but to my surprise it announced mate at 1:59 !
>Sacrificing the pawn by g5-g6 allows an eventual check on f8
>which permits the ensuing mate. Nice problem.
GO for the PENTIUM PRO
Rebel 8.0 (PP-200) (Hash = 13 Mb)
0:39 g6 Mate in 14 moves
- Ed Schroder -
486-133 = 1:59
PRO-200 = 0:39
Howard Exner
Robert Hyatt <hy...@crafty.cis.uab.edu> wrote in article
<52peu1$6...@juniper.cis.uab.edu>...
> Howard Exner (hex...@dlcwest.com) wrote:
> :
> :
> : Emmanuel Marin <Emmanue...@wanadoo.fr> wrote in article
> : <52nq4j$q...@cyan.wanadoo.fr>...
> : > I am curious to know whether chess programs can find the only
> : > correct move for White in this position. ...SNIP...
> : >
> : > White to play and win :
> : >
> : > White : Kc7, Pa5, c2, d5, e5, g5
> : > Black : Ka8, Pa6, b3, c4, e6, f7, h3
>
> I'm assuming g6 is the correct move?
>
> Bob
After 1.g6 fxg6 the d pawn Queens with check then proceeds to
check in this order d4,e4,e3,f3,f2,f8(the point of g6),c5,c6,b6.
The black king shuttles back and forth between a7 and a8. The
zigzag checks leave a nice impression. Because this is 32 ply
does this mean that some knowledge kicked in to find this solution?
Enrique Irazoqui
I have tried this position on a P75 with 13 M. hashtables and Rebel 8 took
2:19 to play g6 at ply 11.04. The strange thing is that under Windows 95
Rebel 8 reached ply 11.04 and indicated g5-g6+ in 1:06, but it didn't play
it and went on to ply 12, 13 and so on. I tried with two computers and
different parameters (W, W4), and still the same story.
Anyway, the P6-200 is 3.5 times faster than a P75. Not bad.
Howard Exner
Just for fun I lowered the hash settings for this problem from 13MB
to 4MB. With 13MB it found the mate in 1:59 but with 4MB Hash it
could not be solved even after 20:00. Never realized that hash size made
such a difference. Also reveals that DOS is the way to go for Rebel 8.0
since 4MB hash is the windows max.
Robert Hyatt
Enrique Irazoqui (en...@bitmailer.net) wrote:
: I have tried this position on a P75 with 13 M. hashtables and Rebel 8 took
Depends on the program, too. For Crafty, the P6/200 is 2.5X faster than a
*good* P5/133... which is almost 2x faster than a p5/75, although for some
P5/75's, my gateway P5/133 is nearly 3x faster... My p5/75 notebook by
Toshiba being a case in point that is 1/3 the speed of the Gateway P5/133 I
used to use on ICC...
Ed Schroder
() wrote:
>"Ed Schröder" <rebc...@xs4all.nl> wrote:
>
>>"Howard Exner" <hex...@dlcwest.com> wrote:
>>>
>>>
>>>Emmanuel Marin <Emmanue...@wanadoo.fr> wrote in article
>>><52nq4j$q...@cyan.wanadoo.fr>...
>>>> I am curious to know whether chess programs can find the only
>>>> correct move for White in this position. ...SNIP...
>>>>
>>>> White to play and win :
>>>>
>>>> White : Kc7, Pa5, c2, d5, e5, g5
>>>> Black : Ka8, Pa6, b3, c4, e6, f7, h3
>>>
>>>CM4000 and MCPro 3.5 were unable to solve this by the 20
>>>minute mark so I stopped the search. I didn't expect Rebel 8.0
>>>to find it either but to my surprise it announced mate at 1:59 !
>>>Sacrificing the pawn by g5-g6 allows an eventual check on f8
>>>which permits the ensuing mate. Nice problem.
>
>
>>GO for the PENTIUM PRO
>
>
>>Rebel 8.0 (PP-200) (Hash = 13 Mb)
>
>>0:39 g6 Mate in 14 moves
>
>>- Ed Schroder -
>
>>486-133 = 1:59
>>PRO-200 = 0:39
>
>
>until the 10:00 mark where I stopped the search... :-(
>(P90, 12 ply)
>
>It seems to me the best pick would be to buy both MCP5 (or 6) and
>Rebel 8.0,
>and to combine their results (we´ ve had cases where MCP finds the
>proper combination and Rebel doesn´t and vice versa, like in this
>case).
>
>Shep
One position does not tell you much, neither does 5-10 games.
20 games will tell you something.
100 games tells you more.
400 - 500 as on SSDF is more than sufficient, despite of all the doubles.
- Ed Schroder -
Robert Hyatt
Ed Schroder (rebc...@xs4all.nl) wrote:
:
:
: One position does not tell you much, neither does 5-10 games.
:
: 20 games will tell you something.
:
: 100 games tells you more.
:
: 400 - 500 as on SSDF is more than sufficient, despite of all the doubles.
:
: - Ed Schroder -
:
They tell you some things, but not everything. For example, how about
500 games against an IM that is about as tactical as anyone I've ever
seen (IMOrlov on ICC) and who *loves* to attack the king. If your king
safety is lacking, it'll show up quickly because you get attacked *every*
game. Most programs (WchessX is a notable exception) are lousy at attacking,
which means that this particular part of most programs is not exposed to
a bright light on the SSDF list. If you'd been on ICC for the past year
or so, you'd have caught Bruce and I (and many others) discussing king
safety and ways to solve the "vicious attack syndrome".
Then you play 500 games against someone like Roman, and you find out how
good your endgame skills are. He usually wades thru the tactics and
reaches endgames that offer substantial challenges to the computer
opponents.
In short, SSDF shows a lot about program vs program, but there's still
another whole world of testing out there, when you let strong humans
beat on it, and beat on it, and beat on it... book learning is almost
mandatory. Playing *strong* is a necessity. :)
Bob
Glenn Rhoads
hy...@crafty.cis.uab.edu (Robert Hyatt) writes:
>: > I am curious to know whether chess programs can find the only
>: > correct move for White in this position. ...SNIP...
>: >
>: > White to play and win :
>: >
>: > White : Kc7, Pa5, c2, d5, e5, g5
>: > Black : Ka8, Pa6, b3, c4, e6, f7, h3
>:
>: I tried this out on my three chess programs on a 486-133.
>: CM4000 and MCPro 3.5 were unable to solve this by the 20
>: minute mark so I stopped the search. I didn't expect Rebel 8.0
>: to find it either but to my surprise it announced mate at 1:59 !
>: Sacrificing the pawn by g5-g6 allows an eventual check on f8
>: which permits the ensuing mate. Nice problem.
>I ran it on crafty 10.15, and at about 10 minutes it went from a
>draw by repetition to g6 with a fail high. I waited another 10
>minutes but got no PV, got disgusted and quit. :) However, it
>would likely play g6 given 10 minutes since it failed high right
>around there. With a larger hash it might be quicker, I didn't
>try anything big...
>I'm assuming g6 is the correct move?
Good question! I did it by hand and found the following forced mate
starting with d6!
1.d6 b2 2.d7 b1=Q 3.d8=Q+ Ka7 4.Qd4+ Ka8 5.Qd5+ Ka7 6.Qc5+ Ka8 7.Qc6+ Ka7
8.Qb6+ Qxb6+ 9.axb6+ Ka8 10.b7+ Ka7 11.b8=Q++
1.g6 still leads to mate though it is one move (two ply) longer.
1.g6 fxg6 2.d6 and wins as before. (1... b2 allows 2.gxf7 b1=Q 3.f8=Q+ Ka7
4.Qc5+ and wins as before except the mate is two moves/4 ply longer.)
Note that in all lines black must queen on b1 in order to stop white from
playing Qb8++.
So to answer the question; g6 is a winning move but it is not the best
move. The original post claimed that there was ONLY ONE correct move.
In some sense both d6 and g6 are correct moves since they both lead to
forced wins, but d6 is better because it wins quicker.
-- Glenn Rhoads
Glenn Rhoads
rho...@paul.rutgers.edu (Glenn Rhoads) writes:
>Good question! I did it by hand and found the following forced mate
>starting with d6!
>1.d6 b2 2.d7 b1=Q 3.d8=Q+ Ka7 4.Qd4+ Ka8 5.Qd5+ Ka7 6.Qc5+ Ka8 7.Qc6+ Ka7
>8.Qb6+ Qxb6+ 9.axb6+ Ka8 10.b7+ Ka7 11.b8=Q++
>1.g6 still leads to mate though it is one move (two ply) longer.
>1.g6 fxg6 2.d6 and wins as before. (1... b2 allows 2.gxf7 b1=Q 3.f8=Q+ Ka7
>4.Qc5+ and wins as before except the mate is two moves/4 ply longer.)
>Note that in all lines black must queen on b1 in order to stop white from
>playing Qb8++.
Someone claimed that REBEL (on a pentium pro) found a mate in 15 beginning
with g6. Is REBEL missing the shorter mate (perhaps it doesn't do the
necessary extensions for this line) or is there some subtle defense to the
line given that I'm missing. What do the various computer programs say is
the best defense to 1.d6, 2.d7, and 3.d8=Q+ ?
-- Glenn Rhoads
john quill taylor
White to play and win :
White : Kc7, Pa5, c2, d5, e5, g5
Black : Ka8, Pa6, b3, c4, e6, f7, h3
M-Chess Pro 5.0 had real trouble with this position,
even through completion of 14-ply using maximum hash
tables! I terminated the search at 100 Million nodes,
since my lunch only lasts 45 minutes ;-).
After forcing 1.g6!! fg 2.d6 b2 3.d7 bQ 4.dQ Ka7,
M-Chess announced a Mate-in-12.
I am guessing that M-Chess would have found 1.g6!! at
15-ply, but that would take several hours on a P6/200,
and probably a week on a 486.
Even using "shortest mate" was of no help at 18-ply.
Initial Position
8 k . . . . . . . (W)
7 . . K . . p . .
6 p . . . p . . .
5 P . . P P . P .
4 . . p . . . . .
3 . p . . . . . p
2 . . P . . . . .
1 . . . . . . . .
a b c d e f g h
jqt M-Chess Pro 5.0
1. g5-g6!! f7xg6 + 0.00
2. d5-d6 b3-b2 + 0.00
3. d6-d7 b2-b1=Q + 0.00
4. d7-d8=Q+ Ka8-a7 +99.87
5. Qd8-d4+ Ka7-a8 +99.88
6. Qd4-e4+ Ka8-a7 +99.89
7. Qe4-e3+ Ka7-a8 +99.90
8. Qe3-f3+ Ka8-a7 +99.91
9. Qf3-f2+ Ka7-a8 +99.92
10. Qf2-f8+ Ka8-a7 +99.93
11. Qf8-c5+ Ka7-a8 +99.94
12. Qc5-c6+ Ka8-a7 +99.95
13. Qc6-b6+ Qb1xb6+ +99.96
14. a5xb6+ Ka7-a8 +99.97
15. b6-b7+ Ka8-a7 +99.98
16. b7-b8=Q+ mate
----------------------------------
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>Howard Exner (hex...@dlcwest.com) wrote:
>:
>:
>: Emmanuel Marin <Emmanue...@wanadoo.fr> wrote in article
>: <52nq4j$q...@cyan.wanadoo.fr>...
>: > I am curious to know whether chess programs can find the only
>: > correct move for White in this position. ...SNIP...
>: >
>: > White to play and win :
>: >
>: > White : Kc7, Pa5, c2, d5, e5, g5
>: > Black : Ka8, Pa6, b3, c4, e6, f7, h3
>:
>: I tried this out on my three chess programs on a 486-133.
>: CM4000 and MCPro 3.5 were unable to solve this by the 20
>: minute mark so I stopped the search. I didn't expect Rebel 8.0
>: to find it either but to my surprise it announced mate at 1:59 !
>: Sacrificing the pawn by g5-g6 allows an eventual check on f8
>: which permits the ensuing mate. Nice problem.
>I ran it on crafty 10.15, and at about 10 minutes it went from a
>draw by repetition to g6 with a fail high. I waited another 10
>minutes but got no PV, got disgusted and quit. :) However, it
>would likely play g6 given 10 minutes since it failed high right
>around there. With a larger hash it might be quicker, I didn't
>try anything big...
>I'm assuming g6 is the correct move?
>Bob
_________________________________________________________________
john quill taylor jqta...@hpdmd48.boi.hp.com writer at large
-------------*-------------
Emmanuel Marin
rho...@paul.rutgers.edu (Glenn Rhoads) wrote:
>Good question! I did it by hand and found the following forced mate
>starting with d6!
>1.d6 b2 2.d7 b1=Q 3.d8=Q+ Ka7 4.Qd4+ Ka8 5.Qd5+ Ka7 6.Qc5+ Ka8 7.Qc6+ Ka7
>8.Qb6+ Qxb6+ 9.axb6+ Ka8 10.b7+ Ka7 11.b8=Q++
Well, after 5.. Ka7 it is true the mate is forced.
But you were so busy looking at the checks from the White Queen you
did not notice that Black can play (5. Qd5+ ???) 5. .. exd5, which
looses the Queen with absolutely no compensation.
Also please note that "the original poster" (wiz : me) only posted
a study made by someone else.
Emmanuel Marin
Paris, France
Joost de Heer
In <52s4m0$7...@paul.rutgers.edu> rho...@paul.rutgers.edu (Glenn Rhoads) writes:
>>: > White : Kc7, Pa5, c2, d5, e5, g5
>>: > Black : Ka8, Pa6, b3, c4, e6, f7, h3
>1.d6 b2 2.d7 b1=Q 3.d8=Q+ Ka7 4.Qd4+ Ka8 5.Qd5+ Ka7 6.Qc5+ Ka8 7.Qc6+ Ka7
>8.Qb6+ Qxb6+ 9.axb6+ Ka8 10.b7+ Ka7 11.b8=Q++
I know it's very very greedy, but I prefer to capture the queen with my e6
pawn, being black, on move 5, instead of getting myself mated.
The 'correct' solution is:
1. g6 fxg6 2. d6 b2 3. d7 b1=Q 4. d8=Q Ka7 5. Qd4 Ka8 6. Qe4 Ka7 7. Qe3 Ka8
8. Qf3 Ka7 9. Qf2 Ka8 10. Qf8 Ka7 11. Qc5 Ka8 12. Qc6 Ka7 13. Qb6 Qxb6 14. axb6
Ka8 15. b7 Ka7 16. b8=Q.
But Ed Schroder said Rebel-8 announced a mate in _14_. Really want to see that
sulution.
>-- Glenn Rhoads
Joost
PS This solution is hand-found, without (ab)use of a computer ;-)
--
Think about all the good in your life - It's only temporary
Think about all the positive sides in life - They never last forever
So drink to forget and drown all your sorrow SENTENCED
Bury your dreams and choose Catharsis NEPENTHE
Joost de Heer
>0:39 g6 Mate in 14 moves
Hmmm. Mate in 14? The original problem was stated as 'mate in 16'. Not only
did Rebel solve the problem, but it cooked it too :-) Gratz, Ed.
>- Ed Schroder -
Joost
Ed Schröder
From: "Howard Exner" <hex...@dlcwest.com>
:Just for fun I lowered the hash settings for this problem from 13MB
About "since 4 Mb hash is the windows max"...
Yes and no.
From a previous discussion here I learned the following:
Win95 (16 Mb) Hash 4-6 Mb is the maximum (depends per program)
Win95 (32 Mb) Hash 20 Mb should give no swapping problems.
Win95 (64 Mb) Hash 32 Mb or more should give no swapping problems.
- REBEL W The hash table is limited to 256 Kb (= REBELWIN setting)
- REBEL W1 The hash table is limited to 512 Kb
- REBEL W2 The hash table is limited to 1 Mb
- REBEL W3 The hash table is limited to 2 Mb
- REBEL W4 The hash table is limited to 4 Mb
I wish I had known this information before and I had extended the
above list for Rebel8.
For now (in case you want the most) run it on DOS (max is 60 Mb hash)
- Ed Schroder -
Simon Read
"Howard Exner" <hex...@dlcwest.com> wrote:
>
>
>Just for fun I lowered the hash settings for this problem from 13MB
>to 4MB. With 13MB it found the mate in 1:59 but with 4MB Hash it
>could not be solved even after 20:00. Never realized that hash size made
>such a difference. Also reveals that DOS is the way to go for Rebel 8.0
>since 4MB hash is the windows max.
What !?!???
This is something I never knew about windows. You're saying that when
M$ "upgraded" the computers from DOS to windows, they REDUCED the
amount of memory the programmer was allowed to use?
Simon
Robert Hyatt
wrote:
: "Howard Exner" <hex...@dlcwest.com> wrote:
:
:
:
: >Just for fun I lowered the hash settings for this problem from 13MB
: >to 4MB. With 13MB it found the mate in 1:59 but with 4MB Hash it
: >could not be solved even after 20:00. Never realized that hash size made
: >such a difference. Also reveals that DOS is the way to go for Rebel 8.0
: >since 4MB hash is the windows max.
: Take a look at the rating list produced by the Louguet (sp?) II Test.
: Increasing the hash table size from around 1MB to 4MB increased the
: performance by about 40 points (MChess 5 on the same machine), if I
: remember correctly. And if you look at the top scores, they were
: reached by programs which permit 64MB hash table size instead of the
: poor 10MB maximum MChess offers.
:
: Shep
:
:
Then you'll love Crafty. :) I have run it on a Cray C90, with 12gb for the
hash table. Yes, that's *gigabytes* on a machine with 16gb of main memory.
Just kidding about the "love crafty" of course, but not about the huge hash
table. Last tournament Cray Blitz played in (that we had access to a C90
for) we used a hash table this big too. I doubt that it helps once the
hash table has more entries than you can hope to search in a given period of
time...
For PC's, bigger is better, because big is not really *big* -yet-...
Robert Hyatt
wrote:
: hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
:
: Is your P6 from Gateway, too? Just curious because I'm considering to
: buy one there...
:
: Shep
:
:
Yes. However, in fairness, I have yet to see a P6/200 that's bad. Main
thing to look out for is the memory. EDO is preferred, fast page mode is
second choice. the 512kb cache option for the newer P6/200 is also a win.
Robert Hyatt
Simon Read (s.r...@cranfield.ac.uk) wrote:
: "Howard Exner" <hex...@dlcwest.com> wrote:
: >
: >
: >Just for fun I lowered the hash settings for this problem from 13MB
: >to 4MB. With 13MB it found the mate in 1:59 but with 4MB Hash it
: >could not be solved even after 20:00. Never realized that hash size made
: >such a difference. Also reveals that DOS is the way to go for Rebel 8.0
: >since 4MB hash is the windows max.
:
: This is something I never knew about windows. You're saying that when
: M$ "upgraded" the computers from DOS to windows, they REDUCED the
: amount of memory the programmer was allowed to use?
:
: Simon
Not really, they just try to allocate real memory in such a way that multiple
programs can co-exist together. If you overstep your system max, and the
program tries to get larger, parts are written to disk, killing performance.
This should be tunable, but I'm not a windows expert...
Enrique Irazoqui
No, it did not cook it. It is a mate in 16 and if you follow Rebel's moves
you will find out. By the way, Nimzo 3 announces mate in 11 after 4:10 on a
P75, which is obviously wrong too. CM5000 gets a mate evaluation (99.84)
after 13 seconds on a P75 and 4 M. tables.
Enrique Irazoqui
Joost de Heer <joo...@sci.kun.nl> wrote in article
<52t6gv$7...@studs2.sci.kun.nl>...
Glenn Rhoads
rho...@crayola.rutgers.edu (Glenn Rhoads) writes:
>>Good question! I did it by hand and found the following forced mate
>>starting with d6!
>>1.d6 b2 2.d7 b1=Q 3.d8=Q+ Ka7 4.Qd4+ Ka8 5.Qd5+ Ka7 6.Qc5+ Ka8 7.Qc6+ Ka7
>>8.Qb6+ Qxb6+ 9.axb6+ Ka8 10.b7+ Ka7 11.b8=Q++
>>1.g6 still leads to mate though it is one move (two ply) longer.
>>1.g6 fxg6 2.d6 and wins as before. (1... b2 allows 2.gxf7 b1=Q 3.f8=Q+ Ka7
>>4.Qc5+ and wins as before except the mate is two moves/4 ply longer.)
>>Note that in all lines black must queen on b1 in order to stop white from
>>playing Qb8++.
>Someone claimed that REBEL (on a pentium pro) found a mate in 15 beginning
>with g6. Is REBEL missing the shorter mate (perhaps it doesn't do the
>necessary extensions for this line) or is there some subtle defense to the
>line given that I'm missing. What do the various computer programs say is
>the best defense to 1.d6, 2.d7, and 3.d8=Q+ ?
Oops! I missed 5.Qd5?? exd5! Now I see the reason for 1.g6.
1.g6 fxg6 (as mentioned above 1... b2 loses to 2.gxf7 b1=Q 3.f8=Q+ Ka7 4.Qc5+
and mates as in the first line listed above)
2.d6 b2 3.d7 b1=Q 4.d8=Q+ Ka7 5.Qd4+ Ka8 6.Qe4+ Ka7 7.Qe3+ Ka8 8.Qf3+ Ka7
9.Qf2+ Ka8 and now the initial sacrifice of the g6 pawn allows
10.Qf8+ Ka7 11.Qc5+ and wins as in the first line.
So it appears that 1.g6 is the unique winning first move after all.
Nice problem!
-- Glenn Rhoads
Joe Stella
"Enrique Irazoqui" <en...@bitmailer.net> wrote:
>[...]
>CM5000 gets a mate evaluation (99.84)
>after 13 seconds on a P75 and 4 M. tables.
>Enrique Irazoqui
CM5000 obviously has a special search extension for checks that have
only one legal reply. This allows it to find very long mates in only
seconds (even when running on "slower" machines) when the mate contains
a series of one-reply checks. This would also explain how it could
miss a shorter mate (if one is present) when the shorter mate contains
checks that have *more* than one legal reply.
Joe Stella
john quill taylor
hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>Depends on the program, too. For Crafty, the P6/200 is 2.5X faster than a
>*good* P5/133... which is almost 2x faster than a p5/75, although for some
>P5/75's, my gateway P5/133 is nearly 3x faster... My p5/75 notebook by
>Toshiba being a case in point that is 1/3 the speed of the Gateway P5/133 I
>used to use on ICC...
This position we are discussing:
White to play and win...
White : Kc7, Pa5, c2, d5, e5, g5
Black : Ka8, Pa6, b3, c4, e6, f7, h3
might be a good one to compare the P200 and 6/200 machines.
It appears that I will have to let my Pentium Pro run for
several hours to "solve" this mate-in-16 with M-Chess Pro 5.0,
but I will do it soon.
Robert: How do I test Crafty on my HP Vectra 6/200? I have to
use a DOS boot disk to run M-Chess; does Crafty run under Win NT,
and if so, where is your program sold? I think I will have to
call ICD and get Rebel 8 soon -- and I am anxiously awaiting
M-Chess 6!
__
john quill taylor / /\
writer at large / / \
Hewlett-Packard, Storage Systems Division __ /_/ /\ \
Boise, Idaho U.S.A. /_/\ __\ \ \_\ \
e-mail: jqta...@hpdmd48.boi.hp.com \ \ \/ /\\ \ \/ /
Telephone: (208) 396-2328 (MDT = GMT - 6) \ \ \/ \\ \ /
Snail Mail: Hewlett-Packard \ \ /\ \\ \ \
11413 Chinden Blvd \ \ \ \ \\ \ \
Boise, Idaho 83714 \ \ \_\/ \ \ \
Mailstop 852 \ \ \ \_\/
\_\/
"When in doubt, do as doubters do." - jqt -
haiti, rwanda, cuba, bosnia, ... we have a list,
where is our schindler?
Robert Hyatt
john quill taylor (jqta...@hpdmd48.boi.hp.com) wrote:
: hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
:
: >Depends on the program, too. For Crafty, the P6/200 is 2.5X faster than a
: >*good* P5/133... which is almost 2x faster than a p5/75, although for some
: >P5/75's, my gateway P5/133 is nearly 3x faster... My p5/75 notebook by
: >Toshiba being a case in point that is 1/3 the speed of the Gateway P5/133 I
: >used to use on ICC...
:
: This position we are discussing:
:
: White to play and win...
:
: White : Kc7, Pa5, c2, d5, e5, g5
: Black : Ka8, Pa6, b3, c4, e6, f7, h3
:
: might be a good one to compare the P200 and 6/200 machines.
:
: It appears that I will have to let my Pentium Pro run for
: several hours to "solve" this mate-in-16 with M-Chess Pro 5.0,
: but I will do it soon.
:
: Robert: How do I test Crafty on my HP Vectra 6/200? I have to
: use a DOS boot disk to run M-Chess; does Crafty run under Win NT,
: and if so, where is your program sold? I think I will have to
: call ICD and get Rebel 8 soon -- and I am anxiously awaiting
: M-Chess 6!
:
: __
It should (according to Bruce Moreland) run under NT with no problems. The
executables are on the machine ftp.cis.uab.edu, anonymous login, directory
is pub/hyatt. source and book stuff is also there. You don't buy it,
you just download it. :)
Don Getkey
In article <52peu1$6...@juniper.cis.uab.edu>, hy...@crafty.cis.uab.edu
(Robert Hyatt) writes:
>: > White : Kc7, Pa5, c2, d5, e5, g5
>: > Black : Ka8, Pa6, b3, c4, e6, f7, h3
>:
>: I tried this out on my three chess programs on a 486-133.
>: CM4000 and MCPro 3.5 were unable to solve this by the 20
>: minute mark so I stopped the search. I didn't expect Rebel 8.0
>: to find it either but to my surprise it announced mate at 1:59 !
>: Sacrificing the pawn by g5-g6 allows an eventual check on f8
>: which permits the ensuing mate. Nice problem.
>
>I ran it on crafty 10.15, and at about 10 minutes it went from a
>draw by repetition to g6 with a fail high. I waited another 10
>minutes but got no PV, got disgusted and quit. :) However, it
>would likely play g6 given 10 minutes since it failed high right
>around there. With a larger hash it might be quicker, I didn't
>try anything big...
>
>I'm assuming g6 is the correct move?
>
>Bob
Curious to see if CM4000/Win95 version could find g6, but while set on 1
move per 3 hours he played d6 in 1:39:01, 12 ply, score 0.00, even though
he was considering g6. This on a P75. :-(
chessman
I don't know about cheating...but what's the point in using one version in
SSDF and another that you sell..I hope no advertiser will try to make the
claim the Genius 4.0 is now xxxx ELO (ugh! ..elo is a whole different
story) on the SSDF...since no one can buy the program ...this whole SSDF
thing is getting out of hand ..it's obvious the SSDF are in the same bed
as the programmers ..it's becoming a farce.....what we really need is group
that has actually purchased the product on the open market ..as it stand
right now, what assurance do we have that the version being tested is the
same as being sold ...opening book and everything the same ...in a word
..NONE. ...this is just my perception...if I'm wrong ..please correct
me...but as they say ..perception is reality.
----------
> From: James Garner <da...@laraby.tiac.net>
> Newsgroups: rec.games.chess.computer
> Subject: Genius Programmer has Few Ethics
> Date: Wednesday, October 02, 1996 8:09 PM
>
> : >Just for fun I lowered the hash settings for this problem from 13MB
> : >to 4MB. With 13MB it found the mate in 1:59 but with 4MB Hash it
> : >could not be solved even after 20:00. Never realized that hash size
made
> : >such a difference. Also reveals that DOS is the way to go for Rebel
8.0
> : >since 4MB hash is the windows max.
>
> That's why Genius cheated by providing a DOS version to the Swiss
> for rating purposes, but is only selling a Windows version to the
> gullible public.
>
Torstein Hall
Chessmaster 5000 gave 1.g6 and mate in 15 in about 10 seconds!
PC= Pentium 100 / 24mb RAM
Perhaps this says someting about CM5K being a very good chessprogram?
--
Torstein Hall (tors...@eunet.no)
Chess page: http://login.eunet.no/~torshall/sjakk.html
Homepage: http://login.eunet.no/~torshall/index.html
Howard Exner <hex...@dlcwest.com> skrev i artikkelen
<01bbaf74$436b6740$2f25...@dlcwest.dlcwest.com>...
>
>
> Robert Hyatt <hy...@crafty.cis.uab.edu> wrote in article
> <52peu1$6...@juniper.cis.uab.edu>...
> > Howard Exner (hex...@dlcwest.com) wrote:
> > :
> > :
> > : Emmanuel Marin <Emmanue...@wanadoo.fr> wrote in article
> > : <52nq4j$q...@cyan.wanadoo.fr>...
> > : > I am curious to know whether chess programs can find the only
> > : > correct move for White in this position. ...SNIP...
> > : >
> > : > White to play and win :
> > : >
> > : > White : Kc7, Pa5, c2, d5, e5, g5
> > : > Black : Ka8, Pa6, b3, c4, e6, f7, h3
> >
> > I'm assuming g6 is the correct move?
> >
> > Bob
>
Enrique Irazoqui
<Unknow wrote in article <52ta3u$p...@news.rrz.uni-koeln.de>...
> hy...@crafty.cis.uab.edu (Robert Hyatt) wrote:
>
> >Enrique Irazoqui (en...@bitmailer.net) wrote:
Hi, answering again for the third time. Something must be wrong with my
news server. Or not. I don't have much of an idea about how to answer here.
So, yes, not all programs increase their computing speed at the same rate
with a faster processor, and my notebooks are also slower than their
corresponding desktops, probably due to the lack of secondary cache.
I did not buy yet a P6. I took as reference the times posted by Ed
Schröder. I am also considering getting a fast notebook: do you know of a
fast one?
Enrique
brucemo
chessman wrote:
>
> I don't know about cheating...but what's the point in using one version in
> SSDF and another that you sell..I hope no advertiser will try to make the
> claim the Genius 4.0 is now xxxx ELO (ugh! ..elo is a whole different
> story) on the SSDF...since no one can buy the program ...this whole SSDF
> thing is getting out of hand ..it's obvious the SSDF are in the same bed
> as the programmers ..it's becoming a farce.....what we really need is group
> that has actually purchased the product on the open market ..as it stand
> right now, what assurance do we have that the version being tested is the
> same as being sold ...opening book and everything the same ...in a word
> ..NONE. ...this is just my perception...if I'm wrong ..please correct
> me...but as they say ..perception is reality.
Perception is reality? Huh?
bruce
Sir Rooke
> >: > I am curious to know whether chess programs can find the only
> >: > correct move for White in this position. ...SNIP...
> >: >
> >: > White to play and win :
> >: >
> >: > White : Kc7, Pa5, c2, d5, e5, g5
> >: > Black : Ka8, Pa6, b3, c4, e6, f7, h3
> >:
> >: CM4000 and MCPro 3.5 were unable to solve this by the 20
> >: minute mark so I stopped the search. I didn't expect Rebel 8.0
> >: to find it either but to my surprise it announced mate at 1:59 !
> >: Sacrificing the pawn by g5-g6 allows an eventual check on f8
> >: which permits the ensuing mate. Nice problem.
>
> >I ran it on crafty 10.15, and at about 10 minutes it went from a
> >draw by repetition to g6 with a fail high. I waited another 10
> >minutes but got no PV, got disgusted and quit. :) However, it
> >would likely play g6 given 10 minutes since it failed high right
> >around there. With a larger hash it might be quicker, I didn't
> >try anything big...
>
I set this problem up on CM5K (P100) and asked for advice. It came up
with "mate in 10" in 20 seconds. I have no idea about my hash tables or
anything else. Are you saying that other programs are taking 20 MINUTES
to do the same???
Ed Schröder
I don't believe this is true!
Why don't you ask SSDF yourself!
SSDF can be found at:
http://www.nsc.liu.se/~bosj/SSDF/
- Ed Schroder -
Dave Gomboc
>> From: James Garner <da...@laraby.tiac.net>
>> Newsgroups: rec.games.chess.computer
>> Subject: Genius Programmer has Few Ethics
>> Date: Wednesday, October 02, 1996 8:09 PM
>>
>> for rating purposes, but is only selling a Windows version to the
>> gullible public.
>>
Give it a rest. The reason Genius' author provided a DOS version is so the
SSDF could auto-test instead of having to operate it manually.
Taking a courteous action and insulting it.. sheesh.
Dave Gomboc
drgo...@a.stu.athabascau.ca
FUCHS ALEXANDER
>Give it a rest. The reason Genius' author provided a DOS version is so the
>SSDF could auto-test instead of having to operate it manually.
>Taking a courteous action and insulting it.. sheesh.
It's ok Richard Lang gave a DOS version of genius to the SSDF. I can
understand they don't like testing manually any more, as autoplayer software
is available.
But, why hasn't the SSDF expressed that clearly in their comments that come
with each list ? When the customer reads "Genius 4" on the Swedish list, of
course he thinks it's the same version that runs on his PC. I've just had a
look at the comments and I could find nowhere something like "to facilitate
out testing, we are using a DOS version of Genius that might be a bit faster
than the commercial windows-version".
I do not intend to speculate about the reasons, but if they had informed
us about this, now there would be no discussion.
Alexander Fuchs
Alexander Fuchs
Robert Hyatt
Sir Rooke (sirr...@ais.net) wrote:
: > >: > I am curious to know whether chess programs can find the only
Yes, but the bad news is, it is not a mate in 10. It's playing the right
move, but it apparently has some sort of bug in the search that doesn't
find the correct defense. I believe it is a mate in 14 or so..
Robert Hyatt
James Garner (da...@laraby.tiac.net) wrote:
: $3306...@omni2.voicenet.com> <325495...@nwlink.com>:
: Distribution:
:
: brucemo (bru...@nwlink.com) wrote:
:
: : > me...but as they say ..perception is reality.
: :
: : Perception is reality? Huh?
:
: If the DOS version were not stronger, why did the Genius
: programmer supply such a version to the SSDF testers?
:
: Perhaps even a more interesting question, why would they treat
: this version as equivalent to the Windows version?
Richard provided this version so it could be connected via the auto232
program, which does ugly things that windows does not allow, like poking
data directly into the console keyboard buffer, and then making the system
think that keys were pressed. Impossible under windows using the same
approach, because that memory is suddenly protected by the 386-up memory
management hardware...
In my testing over the years, I've found no difference between performance in
dos and windows, other than if you make a program too big in windows, it will
start to page, which is bad. The problem is, "what is too big?" Under dos,
it either fits in memory or it doesn't run, unless you use some compiler that
has a built-in pager like the DOS version of gcc...
Torstein Hall
Robert Hyatt <hy...@crafty.cis.uab.edu> skrev i artikkelen
<532trt$v...@juniper.cis.uab.edu>...
My CM5K found mate in 15 in about 10 sec. on a P100 with 24MB ram as posted
earlier. Perhaps the mate in 10 is a human error...? :)
Tord Kallqvist Romstad
James Garner (da...@laraby.tiac.net) wrote:
: $3306...@omni2.voicenet.com> <325495...@nwlink.com>:
: Distribution:
: brucemo (bru...@nwlink.com) wrote:
: : > me...but as they say ..perception is reality.
: :
: : Perception is reality? Huh?
: If the DOS version were not stronger, why did the Genius
: programmer supply such a version to the SSDF testers?
As has been pointed out several times before in this thread, Richard gave
the SSDF a DOS version of Genius 4 to make it possible to autotest the
program. Thanks, Richard!
However, I wish Richard could make the DOS version of G4 commercially
available. I often regret that I upgraded from G3 to G4, because I don't
like the new Windows GUI. I don't think I will upgrade Genius again, if
G5 is also a Windows only program. If a DOS version is released, I would
upgrade at once.
Richard, if you are reading this: Please give us a DOS version!
Tord
brucemo
James Garner wrote:
>
> : >> That's why Genius cheated by providing a DOS version to the Swiss
> : >> for rating purposes, but is only selling a Windows version to the
> : >> gullible public.
>
> : SSDF could auto-test instead of having to operate it manually.
> :
> : Taking a courteous action and insulting it.. sheesh.
>
> been able to "be courteous" and supply a DOS version, in order to get a
> higher rating.
>
> Or perhaps you will now actually try to imply that the Windows and
> the DOS versions would get the same rating. Let's see if you are willing
> to make that statement openly, instead of just letting it sit in an
> implication.
>
> Genius gets to eat his cake and have it too.
You are making the assumption that the program would be higher rated on DOS for some
reason. Richard Lang has responded through an intermediary that he could detect no
speed difference between a Windows version and a DOS version in the past.
What kind of performance problem do you think Windows incurs?
bruce
Komputer Korner
Robert Hyatt wrote:
>
> James Garner (da...@laraby.tiac.net) wrote:
> : $3306...@omni2.voicenet.com> <325495...@nwlink.com>:
> : Distribution:
> :
> : brucemo (bru...@nwlink.com) wrote:
> :
> : : > me...but as they say ..perception is reality.
> : :
> : : Perception is reality? Huh?
> :
> : If the DOS version were not stronger, why did the Genius
> : programmer supply such a version to the SSDF testers?
> : Perhaps even a more interesting question, why would they treat
> : this version as equivalent to the Windows version?
>
> Richard provided this version so it could be connected via the auto232
> program, which does ugly things that windows does not allow, like poking
> data directly into the console keyboard buffer, and then making the system
> think that keys were pressed. Impossible under windows using the same
> approach, because that memory is suddenly protected by the 386-up memory
> management hardware...
>
> In my testing over the years, I've found no difference between performance in
> dos and windows, other than if you make a program too big in windows, it will
> start to page, which is bad. The problem is, "what is too big?" Under dos,
> it either fits in memory or it doesn't run, unless you use some compiler that
> has a built-in pager like the DOS version of gcc...
Has anybody actually bothered to test the difference between Genius DOS and Genius
Windows? CCR articles have said the difference is 4% based on position and mate
testing. Lang and Hyatt swear no difference. Who is right in the case of Genius and
would it matter with any other windows chess programs or would the difference if
any always be the same?
--
Komputer Korner
Komputer Korner
Tord Kallqvist Romstad wrote:
>
> James Garner (da...@laraby.tiac.net) wrote:
> : $3306...@omni2.voicenet.com> <325495...@nwlink.com>:
> : Distribution:
>
> : brucemo (bru...@nwlink.com) wrote:
>
> : : > me...but as they say ..perception is reality.
> : :
> : : Perception is reality? Huh?
>
> : If the DOS version were not stronger, why did the Genius
> : programmer supply such a version to the SSDF testers?
>
> the SSDF a DOS version of Genius 4 to make it possible to autotest the
> program. Thanks, Richard!
>
> However, I wish Richard could make the DOS version of G4 commercially
> available. I often regret that I upgraded from G3 to G4, because I don't
> like the new Windows GUI. I don't think I will upgrade Genius again, if
> G5 is also a Windows only program. If a DOS version is released, I would
> upgrade at once.
>
> Richard, if you are reading this: Please give us a DOS version!
>
> Tord
What is wrong with the Genius Windows GUI?
--
Komputer Korner
Moritz Berger
< snip >
> In my testing over the years, I've found no difference between
performance in
> dos and windows, other than if you make a program too big in windows, it
will
> start to page, which is bad. The problem is, "what is too big?" Under
dos,
> it either fits in memory or it doesn't run, unless you use some compiler
that
> has a built-in pager like the DOS version of gcc...
>
than 5 ELO points) when you run Windows chess programs under Windows
(instead of a DOS version under DOS).
One thing I observed with the paging under Windows: The Win95 paging
algorithm swaps almost everything out of memory with big hash tables but
then stops swapping and doesn't hurt performance (see my post concerning
Fritz 4.01 hash tables). But more than 16 MB RAM is of course recommended
if you're serious about playing strength under Windows. This might also be
a reason why Fritz 4 is rated so low by the SSDF (hint: I have a 40/2h
tournament here between Rebel 8 and Fritz 4 on Pentium 133/166 machines (I
always swapped the machines to make up for speed differences). After 12
games, the standing is 4-2-4! BUT: I have 32 MB hash tables for Fritz, 60
MB hash for Rebel 8. I will post the games and final standings after 20
games (maybe in 2 weeks).
--
-------------
Moritz Berger
ber...@zeus.informatik.uni-bonn.de
Ed Schröder
Komputer Korner <kor...@netcom.ca> wrote:
>Robert Hyatt wrote:
>>
>> James Garner (da...@laraby.tiac.net) wrote:
>> : $3306...@omni2.voicenet.com> <325495...@nwlink.com>:
>> : Distribution:
>> :
>> : brucemo (bru...@nwlink.com) wrote:
>> :
>> : : > me...but as they say ..perception is reality.
>> : :
>> : : Perception is reality? Huh?
>> :
>> : If the DOS version were not stronger, why did the Genius
>> : programmer supply such a version to the SSDF testers?
>> : Perhaps even a more interesting question, why would they treat
>> : this version as equivalent to the Windows version?
>>
>> Richard provided this version so it could be connected via the auto232
>> program, which does ugly things that windows does not allow, like poking
>> data directly into the console keyboard buffer, and then making the system
>> think that keys were pressed. Impossible under windows using the same
>> approach, because that memory is suddenly protected by the 386-up memory
>> management hardware...
>>
>> dos and windows, other than if you make a program too big in windows, it will
>> start to page, which is bad. The problem is, "what is too big?" Under dos,
>> it either fits in memory or it doesn't run, unless you use some compiler that
>> has a built-in pager like the DOS version of gcc...
>
>
>Windows? CCR articles have said the difference is 4% based on position and mate
>testing. Lang and Hyatt swear no difference. Who is right in the case of Genius and
>would it matter with any other windows chess programs or would the difference if
>any always be the same?
>--
>Komputer Korner
It all depends on the hash table size.
- Ed Schroder -
Robert Hyatt
Komputer Korner (kor...@netcom.ca) wrote:
:
:
: Has anybody actually bothered to test the difference between Genius DOS and Genius
: Windows? CCR articles have said the difference is 4% based on position and mate
: testing. Lang and Hyatt swear no difference. Who is right in the case of Genius and
: would it matter with any other windows chess programs or would the difference if
: any always be the same?
: --
Here's the "scoop"...
1. There is absolutely no difference in executing a cpu-bound process
on DOS or windows. Absolutely none.
2. if your process is on the large side, you can incur a penalty when
windows decides that you've got enough real memory, and that anything
else you do is going to start paging to disk. There's a huge performance
hit there.
3. Dos does no multi-tasking, so there's nothing else to steal cpu
cycles. In windows, anything is possible, from a background task doing
something, to network traffic causing background processing, to any
sort of things. If you are concerned, close every application but the
chess program, and unplug the UTP cable to get yourself off the net.
Finally, don't try to make the hash table too large or you will silently
start to page. In dos this is not an issue since dos doesn't page, although
if you were to compile with gcc, it does have a virtual memory manager
built-in and will page for you quite happily.
However, the bottom line is that for a cpu-bound program, the O/S is not
involved at all, and therefore has no influence in what you are doing.
If you do lots of file I/O, windows is not great. However, if you treat
the two systems the same, and don't flip around with the windows and icons
while the chess engine is running, there should be -0- difference, period.
If your chess program does a whole lot of graphics stuff, like maybe
animating a second hand on a clock or something, that might cost, but
that's the programmer's fault for fooling around rather than window's
fault for allowing it... :)
Bob
Joe Stella
"Torstein Hall" <tors...@login.eunet.no> wrote:
It looks like everyone has ignored my last post so let me try again:
CM5000 obviously has a special search extension for checks that have
only one legal reply. This allows it to find very long mates in only
seconds (even when running on "slower" machines) when the mate contains
a series of one-reply checks. This would also explain how it could
miss a shorter mate (if one is present) when the shorter mate contains
checks that have *more* than one legal reply.
Joe Stella
PS. My CM5000 on a P5-100 also finds mate in 16 in a few seconds
(didn't time it exactly...)
Robert Hyatt
James Garner (da...@laraby.tiac.net) wrote:
: Robert Hyatt (hy...@crafty.cis.uab.edu) wrote:
:
: : In my testing over the years, I've found no difference between performance in
: : dos and windows, other than if you make a program too big in windows, it will
: : start to page, which is bad. The problem is, "what is too big?" Under dos,
: : it either fits in memory or it doesn't run, unless you use some compiler that
: : has a built-in pager like the DOS version of gcc...
: And what about hash tables?
Back up to my first sentence above. If you make it too big, it will page,
and performance will die. You can make it too big by making hash tables
larger of course...
:
: Also, what about the fact that when I run Mchess Pro in a DOS
: window under Win95, it runs about 20% fewer positions per second than in
: straight DOS? (True, perhaps a Windows only version of Mchess might be as
: fast, but I kind of doubt it, ya know? I kind of suspect that it is the
: Windows overhead which is sucking up CPU cycles and making Mchess see
: fewer positions per second).
But there *is no overhead* when a program is computing. The only thing
that happens when you run a compute bound process is that the real time clock
generates interrupts, but this happens whether in dos or windows. There is
simply *nothing* to soak up cycles. If you do *anything* in another window,
move the mouse around, or have an optical mouse that gets into an "oscillation"
mode, you can steal cycles. If Mchess Pro does anything odd with graphics
or whatever, it might change a little, although I don't see how it could run
in plain dos and still use the same executable if the executable has windows
system calls in it.
However, the bottom line is that there is *no* overhead in windows when you have
exactly one compute-bound application open, and you keep your hands off the
keyboard and mouse. What does it have to do except run applications?
:
Don Fong
In article <5341bj$d...@news-central.tiac.net>,
James Garner <da...@laraby.tiac.net> wrote:
| Also, what about the fact that when I run Mchess Pro in a DOS
|window under Win95, it runs about 20% fewer positions per second than in
|straight DOS? (True, perhaps a Windows only version of Mchess might be as
|fast, but I kind of doubt it, ya know? I kind of suspect that it is the
|Windows overhead which is sucking up CPU cycles and making Mchess see
|fewer positions per second).
ya know, i kind of suspect you ought to get your facts
straight -before- making accusations.
--
--- don fong ``i still want the peace dividend''
--
Walter Ravenek
> CM5000 obviously has a special search extension for checks that have
> only one legal reply. This allows it to find very long mates in only
> seconds (even when running on "slower" machines) when the mate contains
> a series of one-reply checks. This would also explain how it could
> miss a shorter mate (if one is present) when the shorter mate contains
> checks that have *more* than one legal reply.
It must be more complicated than that, because black can either move his
king or place the queen between the checking queen and his king. The
latter immedeately leads to mate on b8 and to mate in 2 on b7.
Walter Ravenek
Tord Kallqvist Romstad
Komputer Korner (kor...@netcom.ca) wrote:
: Robert Hyatt wrote:
: >
: > James Garner (da...@laraby.tiac.net) wrote:
: > : $3306...@omni2.voicenet.com> <325495...@nwlink.com>:
: > : Distribution:
: > :
: > : brucemo (bru...@nwlink.com) wrote:
: > :
: > : : > me...but as they say ..perception is reality.
: > : :
: > : : Perception is reality? Huh?
: > :
: > : If the DOS version were not stronger, why did the Genius
: > : programmer supply such a version to the SSDF testers?
: > :
: > : Perhaps even a more interesting question, why would they treat
: > : this version as equivalent to the Windows version?
: >
: > Richard provided this version so it could be connected via the auto232
: > program, which does ugly things that windows does not allow, like poking
: > data directly into the console keyboard buffer, and then making the system
: > think that keys were pressed. Impossible under windows using the same
: > approach, because that memory is suddenly protected by the 386-up memory
: > management hardware...
: > In my testing over the years, I've found no difference between performance in
: > dos and windows, other than if you make a program too big in windows, it will
: > start to page, which is bad. The problem is, "what is too big?" Under dos,
: > it either fits in memory or it doesn't run, unless you use some compiler that
: > has a built-in pager like the DOS version of gcc...
: Has anybody actually bothered to test the difference between Genius DOS and Genius
: Windows? CCR articles have said the difference is 4% based on position and mate
: testing. Lang and Hyatt swear no difference. Who is right in the case of Genius and
: would it matter with any other windows chess programs or would the difference if
: any always be the same?
: --
: Komputer Korner
A 4% speed difference means about 3 rating points.
If the DOS version is measurably stronger (I don't believe it is), it is
because of the bigger hash table.
Tord
Tord Kallqvist Romstad
Komputer Korner (kor...@netcom.ca) wrote:
: Tord Kallqvist Romstad wrote:
: >
: > James Garner (da...@laraby.tiac.net) wrote:
: > : $3306...@omni2.voicenet.com> <325495...@nwlink.com>:
: > : Distribution:
: >
: > : brucemo (bru...@nwlink.com) wrote:
: >
: > : : > me...but as they say ..perception is reality.
: > : :
: > : : Perception is reality? Huh?
: >
: > : If the DOS version were not stronger, why did the Genius
: > : programmer supply such a version to the SSDF testers?
: >
: > the SSDF a DOS version of Genius 4 to make it possible to autotest the
: > program. Thanks, Richard!
: >
: > However, I wish Richard could make the DOS version of G4 commercially
: > available. I often regret that I upgraded from G3 to G4, because I don't
: > like the new Windows GUI. I don't think I will upgrade Genius again, if
: > G5 is also a Windows only program. If a DOS version is released, I would
: > upgrade at once.
: >
: > Richard, if you are reading this: Please give us a DOS version!
: >
: > Tord
: What is wrong with the Genius Windows GUI?
I guess it is just a matter of taste, I just liked the look of G3
better. Also, I rarely use any Windows programs except G4. I don't like
to start Windows just to play a game of chess.
I don't understand why Richard couldn't simply sell the MS-DOS and Windows
versions of his program together. Since he already has a nice DOS GUI, this
would require no extra work at all, and I am sure he would sell more copies
of Genius if he did this. DOS and Windows programs both have their advantages,
and I think many people would like to have both.
Tord
: --
: Komputer Korner
Robert Hyatt
Walter Ravenek (rav...@chem.vu.nl) wrote:
I agree, because Crafty also uses the single-respons-to-check
extension. It extends 1 ply when checking the opponent, and an
extra ply if the opposing side has only one move to escape the
check. I think CM5K is doing something different, and better...
Joe Stella
rav...@chem.vu.nl (Walter Ravenek) wrote:
>In article <534360$e...@decius.ultra.net>, jo...@ultranet.com (Joe Stella) wrote:
>> CM5000 obviously has a special search extension for checks that have
>> only one legal reply.
>>[...]
>It must be more complicated than that, because black can either move his
>king or place the queen between the checking queen and his king. The
>latter immedeately leads to mate on b8 and to mate in 2 on b7.
>Walter Ravenek
OK, I guess I didn't say it right. There is only one reply that doesn't
lead to immediate loss of the king. The king is counted as "lost" if
either an illegal move is made (which exposes it to attack), or a move is
made that leads to mate. The mate after the Queen interposition can be easily
seen without any search extensions at all, and it gets counted the same
as an illegal move.
If a program has a "one legal reply" extension but doesn't check to see if
a short mate exists, then it will miss this mate. That must be what
is happening to some of the other programs that can't find it.
Joe Stella
Mark Rawlings
tor...@ifi.uio.no (Tord Kallqvist Romstad) wrote:
>James Garner (da...@laraby.tiac.net) wrote:
>: $3306...@omni2.voicenet.com> <325495...@nwlink.com>:
snip
>As has been pointed out several times before in this thread, Richard gave
>the SSDF a DOS version of Genius 4 to make it possible to autotest the
>program. Thanks, Richard!
>However, I wish Richard could make the DOS version of G4 commercially
>available. I often regret that I upgraded from G3 to G4, because I don't
>like the new Windows GUI. I don't think I will upgrade Genius again, if
>G5 is also a Windows only program. If a DOS version is released, I would
>upgrade at once.
>Richard, if you are reading this: Please give us a DOS version!
>Tord
I happen to really like the Windows GUI of Genius4. No complaints at
all. I think it would be a good idea if *both* versions were released
to the public to make everyone happy. (Maybe a little more work for
the programmers but, hey, that's life!)
Mark
Mark Rawlings
w3fu...@fsrz2.rz.uni-passau.de (FUCHS ALEXANDER) wrote:
>>Give it a rest. The reason Genius' author provided a DOS version is so the
>>SSDF could auto-test instead of having to operate it manually.
>>Taking a courteous action and insulting it.. sheesh.
>It's ok Richard Lang gave a DOS version of genius to the SSDF. I can
>understand they don't like testing manually any more, as autoplayer software
>is available.
>But, why hasn't the SSDF expressed that clearly in their comments that come
>with each list ? When the customer reads "Genius 4" on the Swedish list, of
>course he thinks it's the same version that runs on his PC. I've just had a
>look at the comments and I could find nowhere something like "to facilitate
>out testing, we are using a DOS version of Genius that might be a bit faster
>than the commercial windows-version".
>I do not intend to speculate about the reasons, but if they had informed
>us about this, now there would be no discussion.
>Alexander Fuchs
I agree 100%. It should have been listed in the comments.
Mark
Komputer Korner
Mark Rawlings wrote:
>snipped
OKAY, to sum up here, it seems that with the same program in DOS, you
can get a 4% performance gain from large hash tables, whereas in
Windows, to prevent the swapping, you have to limit the hash tables to
4Mb. So in the end DOS chess programs are stronger, albeit a small
difference of approximately 4% if the DOS program is run with huge hash
tables significantly above 4Mb. There is obviously a practical limit to
the hash table size depending on the amount of computations that a
program does in the time period. Has anybody done experiments to
equate the hash table size vs performance increase all the way up
to the practical limit?
--
Komputer Korner
Dave Gomboc
In article <532omu$3...@news-central.tiac.net>,
James Garner <da...@laraby.tiac.net> wrote:
>: >> That's why Genius cheated by providing a DOS version to the Swiss
>: >> for rating purposes, but is only selling a Windows version to the
>: >> gullible public.
>
>: SSDF could auto-test instead of having to operate it manually.
>:
>: Taking a courteous action and insulting it.. sheesh.
>
>been able to "be courteous" and supply a DOS version, in order to get a
>higher rating.
>
> Or perhaps you will now actually try to imply that the Windows and
>the DOS versions would get the same rating. Let's see if you are willing
>to make that statement openly, instead of just letting it sit in an
>implication.
>
> Genius gets to eat his cake and have it too.
The rating should be +-5 points. Yippy-kai-yay. It is well worth it to
allow autoplay. As for your "I'm sure many others...[wah]" they are all
free to do so. I am sure that if other Windows-based programs (which cannot
be autotested currently) were to provide SSDF with a DOS-based version which
allowed autotesting they'd be used as well.
My understanding is that Mr. Donninger is working to allow Windows
programs to autoplay each other as well. When that becomes a reality,
it will no longer be useful to provide a "substitute" version.
It is possible that SSDF's list should be asterisked to note the fact,
but if that were interpreted by consumers as "it's not the REAL
rating" then Richard Lang could always just not provide a DOS version.
This would lead to a much _less_ accurate rating, since instead of
after a set time 500 games being played and a margin of error of say
+-30, only 150 games would be played with a margin of error of perhaps
+-80, even if the convergance of the first rating is to a slightly
higher point (i.e. 3 points).
Ideally, Richard Lang would have provided both the DOS and Windows 95
versions to his consumers: people could choose which to use, and there'd
be no arguments. Perhaps he did not want to support the old user
interface (bug fixes, enhancements, etc.) in concurrency with the new one.
Dave Gomboc
drgo...@a.stu.athabascau.ca
chessman
Tord Kallqvist Romstad <tor...@ifi.uio.no> wrote in article
<535jm0$i...@maud.ifi.uio.no>...
> Komputer Korner (kor...@netcom.ca) wrote:
> : Robert Hyatt wrote:
> : >
> : > James Garner (da...@laraby.tiac.net) wrote:
> : > : $3306...@omni2.voicenet.com> <325495...@nwlink.com>:
> : > : Distribution:
> : > :
> : > : brucemo (bru...@nwlink.com) wrote:
> : > :
> : > : : > me...but as they say ..perception is reality.
> : > : :
> : > : : Perception is reality? Huh?
> : > :
> : > : If the DOS version were not stronger, why did the Genius
> : > : programmer supply such a version to the SSDF testers?
> : > :
> : > : Perhaps even a more interesting question, why would they
treat
> : > : this version as equivalent to the Windows version?
> : >
> : > Richard provided this version so it could be connected via the
auto232
> : > program, which does ugly things that windows does not allow, like
poking
> : > data directly into the console keyboard buffer, and then making the
system
> : > think that keys were pressed. Impossible under windows using the
same
> : > approach, because that memory is suddenly protected by the 386-up
memory
> : > management hardware...
> : >
{snip]
One issue that the SSDF has is that it's truly not independent...they are
not buying the chess programs on the open market...thus they are subject to
the honesty of the programmer in providing an a 'bona-fide' commercially
available product. The truly good independent *testing* agencies buy the
product on the open market just as a typical consumer would....until they
do that I will always be skepitical that *every* programmer has provided a
non-doctored version of his program...
why would a programmer do this, for the simple reason that if your
*commercially* available version is different than the SSDF version...it
would be that much harder for your competitors to bookup on your program to
gain an advantage with your SSDF version...
Robert Hyatt
Komputer Korner (kor...@netcom.ca) wrote:
The practical limit is nodes_per_second multiplied by the time to move
in seconds. if you have a good replacement strategy, this is about as
good as you can do. If you have a sloppy replacement strategy, you might
still go bigger, because hashing is not uniformly distributed over the
entire hash table, there will always be holes and hot-spots. I suspect
that you get improved performance up to the point where you don't have
to ever overwrite an entry. In the case of Crafty, which is one of the
few programs that does not "clear" the hash table between moves, even bigger
can help, because it can then carry positions from search to search and
maybe use them again, particularly in endgames. In effect, the bigger the
better, although not in every position. But in general...
Shawn K. Quinn
On 4 Oct 1996 16:42:45 +0200, Tord Kallqvist Romstad <tor...@ifi.uio.no> wrote:
>Richard, if you are reading this: Please give us a DOS version!
Or even an OS/2 version (PM if it must be). There is *no* chess software for
OS/2 besides PMICS, which is only a FICS interface (doesn't even work 100%
correctly on ICC).
SKQ
Robert Hyatt
chessman (ches...@voicenet.com) wrote:
:
:
: Tord Kallqvist Romstad <tor...@ifi.uio.no> wrote in article
Yes, but some of the SSDF testers have gotten pretty dang smart. For example,
they discovered that Rebel 7 had some "position cooks" for some test suite
positions, because they reset the position, but reversed the colors, and found
the scores were different. Ed later explained how/why this happened, and it's
not a big deal to me anyway, but the fact that they discovered it lends some
credibility that they would uncover a different "test" program than the actual
"commercial" version...
TEST...@delphi.com
If you run a video benchmark in Windows vs Dos you suffer a 7% hit in
performance.
Flight simmers have known this for years...and the same holds true for W95.
Rainbow V 1.11 for Delphi - Test Drive
Ed Schröder
From: hy...@crafty.cis.uab.edu (Robert Hyatt)
:Yes, but some of the SSDF testers have gotten pretty dang smart. For
example,
:they discovered that Rebel 7 had some "position cooks" for some test
suite
:positions, because they reset the position, but reversed the colors, and
found
:the scores were different. Ed later explained how/why this happened, and
it's
:not a big deal to me anyway, but the fact that they discovered it lends
some
:credibility that they would uncover a different "test" program than the
actual
:"commercial" version...
This needs some correction...
- The subject was Rebel6 not Rebel7
- It was not SSDF but the guys from the Austrian MODUL magazine.
- Subject "Rebel6 and 3-4 cooked positions in the BT-2630 test".
I explained in detail that nothing was cooked.
- Ed -
Jouni Uski
Don't you know, that mchess has maximum depth limit at 26 ply!
Jouni
Robert Hyatt
TEST...@delphi.com wrote:
: If you run a video benchmark in Windows vs Dos you suffer a 7% hit in
: performance.
:
: Flight simmers have known this for years...and the same holds true for W95.
: Rainbow V 1.11 for Delphi - Test Drive
Right, but that's a graphical problem. Rather then being able to poke
things directly into video memory, you have to go thru windows calls to
accomplish the same task... Chess programs don't do much graphical
stuff while searching, so this is not a problem for them...
Robert Hyatt
Ed Schröder (rebc...@xs4all.nl) wrote:
: From: hy...@crafty.cis.uab.edu (Robert Hyatt)
:
As I said, I wasn't implying that you did. And that I accepted your
explanation of what happened. I had thought it was some of the SSDF
testers that first noted this, but stand corrected. I still stand by
my statement that were they to be given something obviously different
from a commercial version, someone would notice and raise heck.
Ed Schröder
I did not felt attacked...
- Ed -
john quill taylor
df...@cse.ucsc.edu (Don Fong) wrote:
>In article <5341bj$d...@news-central.tiac.net>,
>James Garner <da...@laraby.tiac.net> wrote:
>| Also, what about the fact that when I run Mchess Pro in a DOS
>|window under Win95, it runs about 20% fewer positions per second than in
>|straight DOS? (True, perhaps a Windows only version of Mchess might be as
>|fast, but I kind of doubt it, ya know? I kind of suspect that it is the
>|Windows overhead which is sucking up CPU cycles and making Mchess see
>|fewer positions per second).
> ya know, i kind of suspect you ought to get your facts
>straight -before- making accusations.
My experiences with M-Chess 3.5, 4.0, and 5.0 concur with James'.
It is much slower under Windows 3.1 or Windows 95, even when
optimized. The programmer has even designed special modes to
run under windows, which help a bit, but the best performance is
with plain DOS and with no memory manager (mcp5/x mode).
__
john quill taylor / /\
writer at large / / \
Hewlett-Packard, Storage Systems Division __ /_/ /\ \
Boise, Idaho U.S.A. /_/\ __\ \ \_\ \
e-mail: jqta...@hpdmd48.boi.hp.com \ \ \/ /\\ \ \/ /
Telephone: (208) 396-2328 (MDT = GMT - 6) \ \ \/ \\ \ /
Snail Mail: Hewlett-Packard \ \ /\ \\ \ \
11413 Chinden Blvd \ \ \ \ \\ \ \
Boise, Idaho 83714 \ \ \_\/ \ \ \
Mailstop 852 \ \ \ \_\/
\_\/
"When in doubt, do as doubters do." - jqt -
haiti, rwanda, cuba, bosnia, ... we have a list,
where is our schindler?
Don Fong
In article <53bkat$5...@hpbs2500.boi.hp.com>,
john quill taylor <jqta...@hpdmd48.boi.hp.com> wrote:
>df...@cse.ucsc.edu (Don Fong) wrote:
>
>>In article <5341bj$d...@news-central.tiac.net>,
>>James Garner <da...@laraby.tiac.net> wrote:
>>| Also, what about the fact that when I run Mchess Pro in a DOS
>>|window under Win95, it runs about 20% fewer positions per second than in
>>|straight DOS? (True, perhaps a Windows only version of Mchess might be as
>>|fast, but I kind of doubt it, ya know? I kind of suspect that it is the
>>|Windows overhead which is sucking up CPU cycles and making Mchess see
>>|fewer positions per second).
>
>> ya know, i kind of suspect you ought to get your facts
>>straight -before- making accusations.
>
>My experiences with M-Chess 3.5, 4.0, and 5.0 concur with James'.
>It is much slower under Windows 3.1 or Windows 95, even when
>optimized.
remember though, that windows (at least under win3.1)
does "preemptive multi-tasking" among DOS windows; but it does
-not- do preemptive multi-tasking among windows programs.
windows (3.1) does only "cooperative multi-tasking" among windows
programs. let that sink in... it is not obvious whether the
performance of a DOS program under windows would be as good
as a windows program under windows, or vice versa.
there is also the question of what effect the scheduling
priorities have, and whether they have been tuned optimally for
your test conditions.
that is not to say that i necessarily agree or disagree
with either side's performance claims.
my only point was, and remains:
get your facts straight -before- making accusations.
--
Jouni Uski
>
> My experiences with M-Chess 3.5, 4.0, and 5.0 concur with James'.
> It is much slower under Windows 3.1 or Windows 95, even when
> run under windows, which help a bit, but the best performance is
> with plain DOS and with no memory manager (mcp5/x mode).
>
My own experience with 486 PC: With x- mode Mchess 4 is faster in
positions/s, BUT there is some kind of bug, because solution times
for problems are LONGER then in normal mode!!
Has anybody explanation? I haven't used x-mode after finding this
flaw...
Jouni
Robert Hyatt
Jouni Uski (Jouni...@nce.nokia.com) wrote:
: >
: > My experiences with M-Chess 3.5, 4.0, and 5.0 concur with James'.
It's an anomoly caused by the transposition table. As you make the
table bigger, the program is able to graft parts of the tree from
one section to another. Sometimes this results in searching fewer
nodes and so the search goes faster. Other times, this "grafted"
information causes a branch to not fail high, when it would without
the transposition table hit. This branch then is not the only
one that has to be searched. The trade-off is that the slower
search is actually somewhat deeper than it's being given credit
for, so I'd take it anyway...
Jouni Uski
But I have tested with exactly same hash size in normal and x mode!
Jouni
Robert Hyatt
Jouni Uski (Jouni...@nce.nokia.com) wrote:
In that case, I'm clueless. Just make sure your disk access light is not
on at all, which would indicate paging. Is this Win95? If so, I don't know
that we have any up yet. 3.11 is everywhere however...
Bob
Goran Grottling
>As I said, I wasn't implying that you did. And that I accepted your
>explanation of what happened. I had thought it was some of the SSDF
>testers that first noted this, but stand corrected. I still stand by
>my statement that were they to be given something obviously different
>from a commercial version, someone would notice and raise heck.
>
Yes! You can be sure about that!
Goran G.
Pitters
Im Artikel <53g4hf$c...@juniper.cis.uab.edu>, hy...@crafty.cis.uab.edu
(Robert Hyatt) schreibt:
ChessGenius 4 in the SSDF rating list
There is no doubt that Richard Lang has been world's most successful
computer
chess programmer ever. He has won no less than ten world championship
titles
since 1984. It's not surprising that this amount of success has made some
people jealous.
Recently there were discussions about the performance of ChessGenius 4 in
the
SSDF rating list. As a matter of fact Richard Lang has been asked to
supply a
DOS-version of ChessGenius 4 which is compatible with the autoplayer
designed
by Chrilly Donninger. Richard was friendly enough to fulfil this request.
I
personally find it very unfair that now certain people are blaming him for
being kind.
It's even more unfair that even a fellow programmer brings up the
suspicion
that this would give ChessGenius 4 program an undeserved advantage. This
is
completely untrue!
As a matter of fact we have conducted various tests which haven't showed
any
measurable difference in speed or performance between the DOS and the
Windows
versions of ChessGenius 4. There is absolutely no technical reason why
there
should be any difference, if the Windows software is programmed in the
correct
way. The only theoretical difference lies in the fact that a DOS version
could
use 15 MB of RAM on a 16 MB PC, whereas the Windows version could probably
use
only 13,5 MB. Every expert knows that this difference is completely
insignificant.
But even if there would be a speed difference of 1,0 %, what would that
mean?
If we assume that the speed difference between a 486/50 MHz and a Pentium
90
MHz is round about 100 %, and we see that this makes only 18 Elo points
difference on the Swedish list, then the conclusion is clear:
1 % added speed gives ChessGenius 4
LESS THAN 0,2 (!!) extra Elo points.
This is far below any measurable tolerance, and even 10 % wouldn't mean
anything.
The discussions in the Internet forum have shown to us that there is still
demand for DOS versions of ChessGenius, although the big majority prefers
the
more user-friendly Windows version. In the end of November 1996 we will
bring
out the new ChessGenius 5 in a DUAL VERSION, which means DOS version and
Windows version together on one CD-ROM. This will hopefully satisfy all
critics and allow everybody to make his own evaluations.
After more that 120 test games we are very optimistic about the
performance of
the new beta-version. We believe that it plays ca. 35 - 50 points stronger
than ChessGenius 4.
Ossi Weiner
HCC, Munich
Tord Kallqvist Romstad
Pitters (pit...@aol.com) wrote:
: The discussions in the Internet forum have shown to us that there is still
: demand for DOS versions of ChessGenius, although the big majority prefers
: the
: more user-friendly Windows version. In the end of November 1996 we will
: bring
: out the new ChessGenius 5 in a DUAL VERSION, which means DOS version and
: Windows version together on one CD-ROM. This will hopefully satisfy all
: critics and allow everybody to make his own evaluations.
Great!
Exactly what I was hoping for...
I look forward to upgrading to Genius 5.
Tord
Enrique Irazoqui
James Garner <da...@laraby.tiac.net> wrote in article
<53me5a$d...@news-central.tiac.net>...
> Tord Kallqvist Romstad (tor...@ifi.uio.no) wrote:
> : Great!
> : Exactly what I was hoping for...
>
> Would you have gotten it but for this thread?
Certainly. Any time.
Enrique