Author: Robert Hyatt
Date: 16:29:24 09/04/02
Go up one level in this thread
On September 03, 2002 at 11:55:09, Vincent Diepeveen wrote:
>On September 02, 2002 at 23:42:37, Robert Hyatt wrote:
>
>>On September 02, 2002 at 23:33:06, Robert Hyatt wrote:
>>
>>>
>>>
>>>
>>>
>>>>
>>>>If i compile crafty single cpu and compare it with a dual crafty,
>>>>your speedup is about 1.4, simple as that.
>>
>>What is simple is your methodology. The difference in speed between
>>the non-SMP version and the SMP version is < .1%. The SMP version using
>>one cpu has one extra load and branch than the non-SMP version. In search.c.
>>It is easy to see, and trivial to prove that your statement is a crock...
>
>So you deny that if you split that this gives overhead > 0.1% ?
OK... let's try to nail this down one last time. You said the SMP version
was "hell slower" than the non-SMP version. The only way the two can be
compared whatsoever is with one cpu. The difference there is almost zero.
So, lets try the next angle below:
>
>If you run with crafty at 2 processes, i hope you realize the
>average user hopes he gets in nodes a second 2 times faster than
>when running 1 cpu.
>
>With or without slow memory.
OK... here are my numbers for one cpu, two cpus and three cpus:
time=57.68 cpu=100% mat=0 n=21748257 fh=92% nps=377k
time=34.20 cpu=199% mat=0 n=24293957 fh=91% nps=710k
time=23.41 cpu=298% mat=0 n=24227814 fh=91% nps=1034k
Now, how do those compare: 1 to 2 is 1.88X faster.
1 to 3 is 2.75X faster.
>
>You deny that the single cpu version gets more nodes a second than
>2 processors * 2 ?
>
Do you claim that 1.88x is "hell slower" than the one cpu version?
>
>
>>
>>
>>>>
>>>>Perhaps Nalimov has an explanation for it from compiler viewpoint
>>>>(having to do with a single thread versus 2 threads), i don't know
>>>>it from compiler viewpoint with regard to multithreading,
>>>>but i see a huge speed diff.
>>
The above is the point. YOu say "huge difference". I say 6%. I have
numbers to back up my results. Others can produce the _same_ test numbers
if they want...
>>
>>Eugene won't have any explanation, because there _is_ none. So I have
>>absolutely no idea what you are talking about... you claimed a dual could
>>only produce a 1.0 speedup on the Cray Blitz / Mchess Pro game in the DTS
>>article. Both I and GCP provided data that looked quite natural, mine
>>showing 3.1, which I emailed to you if you remember...
>
>Noop GCP showed 2.8 speedup very clearly. Shall i repost his email
>again?
Did I say GCP said anything about 3.x above? Or did I say _my_ test showed
3.1 (it actually showed 3.0 for null-move on). Can you read? Can you
comprehend? I have said GCP's number was 2.8 _several_ times already.
>
>*every* posting you do here you deny numbers simply.
>
>Do you have such a bad memory?
>
>*you had 1 stupid position where you showed 3.0 under your ideal
>conditions*.
I ran the entire 24 position set, vincent, and you _know_ that...
Not "one position"...
>
>So with a very silly asymmetric evaluation and at your slow hardware
>where cpu's are not much faster than memory!
>
Of course. Handwave.
>>Of course, don't let some real data get in the way of your wild statements...
>>No need to let facts confuse fiction...
>>
>>
>>
>>
>>
>>>>
>>>>If everyone at home can compile crafty and see that if they optimize
>>>>it single cpu they can get it very fast versus dual it is only 1.4
>>>>times faster for them, then that's really a big problem.
>>>>
>>>>In fact you can already see the speed diffs from the simple tests
>>>>that it isn't true that it can ever get 1.7 speedup, because
>>>>it already has effort to get 1.7 times the number of nodes a second,
>>>>which you wisely hide by not printing out the number of nodes a
>>>>second it gets at each ply.
>>>>
>>>>Even when slate at his slow chipset is running crafty single cpu
>>>>he gets nearly a million nodes a second. Dual it gets 1.6MLN nodes
>>>>a second.
>>
>>
>>
>>So? That is a motherboard/chipset issue, not an issue inside Crafty. I
>>have explained this many times. Duals don't have interleaved memory. Quads
>>do. At least _most_ duals don't. Eugene posted data showing that new Intel
>>PIV duals scale very well while the AMD machine he tested was horrible in a
>>dual config.
>>
>>Nothing to do with Crafty... Everything to do with the machine itself...
>>
>>
>>
>>
>>
>>>>
>>>>The really optimized single cpu machines get however way over a million
>>>>nodes a second. That's at say 1.73Ghz. K7 even.
>>>>
>>>>P3 is so outdated we cannot count it anymore. A single cpu K7 is outsearching
>>>>crafty with crafty at internet. We have seen it so many times at the
>>>>internet.
>>>>
>>>>The only 'luck' you have at your quad is that Corbitt or whoever compiles
>>>>your crafty version doesn't know a thing from compiling (i'm not claiming
>>>>i know more, let's be clear here) and produces a version which at
>>>>any K7 or P4 is hell slower than a default msvc compile here.
>>>>
>>>>Let's not talk about special k7 extensions or special P4 extensions
>>>>(at the k7 there is more to win than at P4 IMHO).
>>>>
>>>>If we compare those versions with 100% same compiled versions for k7
>>>>and see its speedup then it's hell slow.
>>>>
>>>>Now you will complain about memory being slow on K7 but that's simply
>>>>not true. It's about 680MB/s that chipset which is very fast.
>>
>>
>>That is a crock. bandwidth is _not_ the issue. The bus is the issue when
>>you have two processors. _everybody_ has seen that the duals are simply not
>>as good, and it is well known and often-discussed on various SMP mailing
>>lists...
>>
>>
>>
>>
>>>>
>>>>The supercomputer chipsets usually do not have more in fact and it is
>>>>not going to change in the future either for quads, unless you
>>>>only are going for hammer which is having other problems though.
>>
>>
>>The quad xeon in my office has roughly 4X the bandwidth of any single or
>>dual you can compare with using a 100mhz FSB. Of course, again don't let
>>any architectural ideas influence your "how it _must_ be, because I say it
>>is that way..."
>>
>>
>>
>>
>>>>
>>>>Your P3 processors are so slow (700Mhz or so?) that you don't see that
>>>>if they would be clocked at 1.6 or 1.7 or even 2.5Ghz, that the memory
>>>>also gets 3 times slower. Parallel or not.
>>
>>So what? with 4-way interleaving the memory is _still_ 4x faster than it
>>would be on any dual or single cpu machine....
>>
>>
>>
>>>>
>>>>You HAVE to do tests with slower memory simply at duals. You can't
>>>>claim it's 1.7 simply based upon the weakest evidence ever.
>>
>>
>>I (and others) report speedup simply by running a position using one cpu,
>>and then running it with two, and computing the speedup by dividing. It is
>>_really_ easy to do, and it factors in _everything_... You like to make
>>excuses about your poor results, by waving hands here and there... (another
>>followup coming, netscape about to crash)
>>
>>
>>
>>
>>
>>>>
>>>>But the real parallel problem is different in crafty.
>>>>
>>>>Crafty without asymetric king safety is having a horrible
>>>>speedup at a dual. In fact i measured about 1.0 - 1.4 when forgetting
>>>>the single cpu versus smp diff.
>>>>
>>>>This is very weird, because only that one feature is the difference
>>>>between normal chess programs.
>>>>
>>>>If a normal chessprogram uses your parallellism it is giving a speedup
>>>>of 1.0 to 1.4, whereas only one stupid feature which is doing so bad
>>>>in computer-computer play for them (because their king safety is better,
>>>>and play against humans is not important simply, as they do not sell
>>>>based upon how well it plays humans at the icc), but it is of course
>>>>great to parallellize a program because it gets like death penalty
>>>>for some things.
>>>>
>>>>That means that searching in parallel will give huge cutoffs both
>>>>sides. That improves branching factor a little, because *still* it
>>>>isn't getting a 1.7 speedup in nodes a second. In fact i measured
>>>>crafty to lose about 30-35% speed when running parallel versus
>>>>single cpu.
>>>>
>>>>So you can according to your own theories (which say it is impossible
>>>>to get a > 2.0 speedup, though i disagree with that) not even get
>>>>a 1.7 speedup, because you
>>>>would need less nodes a ply then.
>>>>
>>>>In fact if you accurately print out how many nodes a ply you need you
>>>>will see that also crafty needs in many cases LESS nodes at bigger
>>>>depths (with the asymetric feature of course), and that only because
>>>>of it's dead slow speed when running dual, you get a 1.4 speedup
>>>>30% loss from 2.0 = 1.4, so at very old machines where memory is
>>>>less of a problem and you suffer less from all that slow parallel
>>>>stuff you lose less than 30%, but say 15% and 15% loss from 2.0 = 1.7
>>>>
>>>>That's why you claim 1.7.
>>>>
>>>>Note that then directly your formula of 1 + 0.7 (n-1) for n cpu's
>>>>is not working, because GCPs test clearly showed it had a speedup
>>>>of 2.8 at your ideal slow hardware (slow cpu's).
>>>>
>>>>In the future RAM won't get that much faster, whereas cpu's will
>>>>double in speed and so on. Obviously you still run at your quad
>>>>xeon in 2010 to maintain a 1.7 speedup and by proof of 1 carefully
>>>>selected position (where even a stupid beta version of diep with
>>>>a very bad parallellism even could achieve a 1.9 speedup at 2
>>>>processors, whereas at normal positions it got a 1.4 speedup
>>>>at most) you show it's 1.7 at your quad xeon at 2 cpu's. Of course
>>>>3 cpu's never get tested nor 4 nor 5 nor 6 nor 7.
>>>>
>>>>Because we both know that at 32 cpu's you'll never get n + 0.7 (n-1),
>>>>even if memory is parallel fast.
>>>>
>>>>Of course no such machines exist, but it's very easy proof to show
>>>>why it won't work.
>>>>
>>>>Each cpu copies 44KB datastructure or something to split...
>>>>
>>>>>On September 01, 2002 at 09:10:56, Vincent Diepeveen wrote:
>>>>>
>>>>>>On August 31, 2002 at 23:43:53, Robert Hyatt wrote:
>>>>>>
>>>>>>Dead wrong Bob, Again your horrible memory is the problem.
>>>>>
>>>>>
>>>>>OK.. I believe you are right here. In fact, it is possible that
>>>>>O-O was a book move. It is in my book for example, and g3 is the
>>>>>next book move in crafty's book.
>>>>>
>>>>>But as far as horrible memory goes, I pointed out _several_ errors _you_
>>>>>made and I noticed that your "horrible responses" were simply missing.
>>>>>
>>>>>you did some lousy math on the DTS numbers, and when I pointed out the
>>>>>error, you disappear. You later claim Crafty gets a speedup of 1.0 on your
>>>>>machine, on the positions used for the DTS article, and then GCP and myself
>>>>>both got results around 3.0... no response.
>>>>>
>>>>>I like the way you respond to what is convenient... And how you continually
>>>>>say that crafty's eval is bad, or its search is bad, but never address the
>>>>>point that you can't win more than 1 of 3 games from it.
>>>>>
>>>>>Which is worse, a memory that does occasionally make an error, as above?
>>>>>Or someone with an ego that makes statements they can't back up, and when
>>>>>challenged, they run and hide until things disappear???
>>>>>
>>>>>I know which _I_ would rather deal with...
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>
>>>>>>Comm line went down before c4, not before o-o. I have eye witness
>>>>>>report from Frans Morsch here. He was sitting 1.5 meters away
>>>>>>from Deep Blue there and of course being opponent he asked
>>>>>>what happened. Comm line did *not* go away at that time. comm line
>>>>>>was excellent.
>>>>>>
>>>>>>They played o-o because the thing knew shit from chess!
>>>>>>
>>>>>>Older versions of diep (1997) need about 8-9 ply to find g3 to be
>>>>>>better. Current version 10 ply which in itself i already consider
>>>>>>bad. They didn't find it at all of course.
>>>>>
>>>>>They did find it, as they reported. But they didn't find it after they
>>>>>restarted...
>>>>>
>>>>>That happens.
>>>>>
>>>>>Perhaps every time you point out an error their program made, I'll take the
>>>>>time to point out _ten_ that yours makes today? I've _personally_ seen your
>>>>>program castle into vicious attacks. All the while thinking it is winning.
>>>>>Programs make mistakes. Deep Thought. Deep Blue. And yours too...
>>>>>
>>>>>>
>>>>>>It is the c4 move which deep blue played nearly instantly. However
>>>>>>DIEP sees instantly that c4 loses, in fact it doesn't consider playing
>>>>>>it even. Nevertheless we know that even with 500k nodes a second older
>>>>>>versions of DT they only can search 8 ply, so it is logical that they
>>>>>>do not see c4 to be losing if they play instantly a move.
>>>>>>
>>>>>
>>>>>Sorry, but I watched deep thought. Cray blitz searched 10 plies in 1992.
>>>>>I don't have many logs left, but I do have our first GM win in a tournament
>>>>>game vs Van Der Sterren... and it was doing 10 plies. And deep thought
>>>>>was going deeper...
>>>>>
>>>>>
>>>>>
>>>>>>Here what diep would play there:
>>>>>>
>>>>>>00:00 100 0k 0 0 5 (2) 1 (0,0) -1.130 Na3xb5 a6xb5 Bd3xb5
>>>>>>00:00 137 0k 0 0 11 (2) 1 (0,0) -0.113 Rf1-g1
>>>>>>00:00 115 0k 0 0 15 (2) 1 (0,0) -0.089 g2-g3
>>>>>>00:00 100 0k 0 0 17 (2) 1 (0,0) -0.067 Qd1-h5
>>>>>>00:00 100 0k 0 0 20 (2) 1 (0,0) -0.055 Bd3-e2
>>>>>>00:00 100 0k 0 0 24 (2) 1 (0,0) -0.008 c2-c3
>>>>>>00:00 322 0k 0 0 100 (2) 2 (0,3) -0.381 c2-c3 Qd8-g5
>>>>>>++ d1-h5
>>>>>>00:00 553 1k 0 0 216 (2) 2 (0,6) -0.081 Qd1-h5 h7-h6
>>>>>>00:00 1217 1k 0 0 499 (2) 3 (0,41) 0.007 Qd1-h5 h7-h6 c2-c3
>>>>>>00:00 6548 7k 0 0 3209 (2) 4 (2,80) -0.409 Qd1-h5 h7-h6 h2-h3 Nc6-d4
>>>>>>++ c2-c3
>>>>>>00:00 6910 7k 0 0 3801 (2) 4 (2,85) -0.375 c2-c3 Qd8-g5 g2-g3 Be6-h3
>>>>>>++ g2-g3
>>>>>>00:00 11971 12k 0 0 7662 (2) 4 (6,99) -0.350 g2-g3 Be6-h3 Rf1-g1 Qd8-g5
>>>>>>++ f1-g1
>>>>>>00:00 12972 13k 0 0 9081 (2) 4 (7,107) -0.294 Rf1-g1 Qd8-g5 Bd3-f1 Nc6-d4
>>>>>>00:00 18457 18k 0 0 15320 (2) 5 (8,169) -0.224 Rf1-g1 Qd8-g5 Bd3-f1 f7-f5 Nd5-
>>>>>>b6
>>>>>>++ c2-c3
>>>>>>00:01 23652 24k 0 0 24362 (2) 5 (9,182) -0.127 c2-c3 Qd8-g5 g2-g3 f7-f5 h2-h4
>>>>>>
>>>>>>00:01 31402 31k 0 0 38939 (2) 6 (12,246) -0.300 c2-c3 Qd8-g5 g2-g3 Be6-g4 Bd3-
>>>>>>e2 Bg4-h3
>>>>>>00:02 60220 60k 0 0 135495 (2) 7 (31,370) -0.287 c2-c3 Qd8-g5 g2-g3 f7-f5 h2-h
>>>>>>4 Qg5-h6 Qd1-f3 f4xg3 f2xg3
>>>>>>++ g2-g3
>>>>>>00:04 74426 74k 0 0 320777 (2) 7 (34,396) -0.210 g2-g3 Nc6-e7 Nd5-b4 Be6-h3 Rf
>>>>>>1-g1 Qd8-b6 Qd1-h5
>>>>>>00:07 78060 78k 0 0 611993 (2) 8 (75,667) -0.401 g2-g3 Be6-g4 Bd3-e2 Bg4xe2 Qd
>>>>>>1xe2 Nc6-d4 Qe2-d1 Qd8-g5
>>>>>>++ c2-c3
>>>>>>00:08 81916 82k 0 0 715949 (2) 8 (81,725) -0.310 c2-c3 Qd8-g5 g2-g3 f7-f5 h2-h
>>>>>>4 Qg5-d8 Qd1-f3 f4xg3 f2xg3
>>>>>>00:26 101208 101k 0 0 2651670 (2) 9 (268,1617) -1.128 c2-c3 Qd8-h4 g2-g3 f4xg3 f
>>>>>>2xg3 Rg6xg3 Qd1-e1 Rg3-h3 Qe1-e2
>>>>>>++ g2-g3
>>>>>>00:48 113156 113k 0 0 5474505 (2) 9 (550,2763) -0.153 g2-g3 Be6-h3 Rf1-g1 f4xg3
>>>>>>f2xg3 Nc6-d4 Bd3-e2 Nd4xe2 Qd1xe2
>>>>>>01:01 117297 117k 0 0 7226727 (2) 10 (626,3357) -0.354 g2-g3 Be6-h3 Rf1-g1 Nc6-d
>>>>>>4 Bd3-e2 Nd4xe2 Qd1xe2 f4xg3 f2xg3 Bf8-h6
>>>>>>
>>>>>>now if i play c4, *after* cleaning hashtables first then starting
>>>>>>search (from blacks viewpoint is the score, so positive means good for
>>>>>>black):
>>>>>>
>>>>>>00:00 100 0k 0 0 5 (2) 1 (0,0) -0.363 b5xc4 Na3xc4
>>>>>>00:00 130 0k 0 0 13 (2) 1 (0,0) -0.026 Be6xd5 c4xd5
>>>>>>00:00 130 0k 0 0 17 (2) 1 (0,0) 0.105 b5-b4
>>>>>>00:00 116 0k 0 0 21 (2) 1 (0,0) 0.497 Qd8-g5
>>>>>>00:00 609 1k 0 0 128 (2) 2 (0,1) 0.277 Qd8-g5 g2-g3
>>>>>>00:00 3544 4k 0 0 957 (2) 3 (1,44) 0.516 Qd8-g5 Rf1-g1 b5-b4
>>>>>>++ c6-d4
>>>>>>00:00 6014 6k 0 0 2105 (2) 3 (1,54) 0.563 Nc6-d4 c4xb5 Qd8-g5
>>>>>>00:00 10014 10k 0 0 4106 (2) 4 (1,91) 0.273 Nc6-d4 c4xb5 Qd8-g5 g2-g3 f4xg3 h2
>>>>>>xg3 Nd4xb5 Bd3xb5 a6xb5 Na3xb5
>>>>>>++ d8-g5
>>>>>>00:00 11314 11k 0 0 6223 (2) 4 (5,100) 0.366 Qd8-g5 g2-g3 Be6-h3 Rf1-g1
>>>>>>00:00 20072 20k 0 0 13248 (2) 5 (9,158) 0.500 Qd8-g5 g2-g3 Nc6-d4 c4xb5 Be6-g4
>>>>>>
>>>>>>00:01 34894 35k 0 0 54086 (2) 6 (19,238) 0.425 Qd8-g5 g2-g3 Nc6-e7 Nd5xe7 Bf8x
>>>>>>e7 c4xb5
>>>>>>00:02 52132 52k 0 0 147535 (2) 7 (24,364) 0.487 Qd8-g5 g2-g3 Nc6-e7 h2-h4 Qg5-
>>>>>>h6 Qd1-f3 Ne7xd5 e4xd5 f4xg3 f2xg3
>>>>>>00:08 67758 68k 0 0 571207 (2) 8 (72,733) 0.402 Qd8-g5 g2-g3 Be6-g4 Bd3-e2 Bg4
>>>>>>xe2 Qd1xe2 Nc6-d4 Qe2-d3 b5xc4 Na3xc4 f4xg3 h2xg3
>>>>>>++ d8-h4
>>>>>>00:18 93952 94k 0 0 1742828 (2) 8 (131,966) 1.670 Qd8-h4 g2-g3 Qh4-h3 Qd1-f3 R
>>>>>>g6-h6 Qf3-g2 Qh3xg2 Kh1xg2 Be6-h3 Kg2-h1 Bh3xf1 Ra1xf1
>>>>>>00:24 103166 103k 0 0 2553361 (2) 9 (134,1070) 1.670 Qd8-h4 g2-g3 Qh4-h3 Qd1-f3
>>>>>>Rg6-h6 Qf3-g2 Qh3xg2 Kh1xg2 Be6-h3 Kg2-h1 Bh3xf1 Ra1xf1
>>>>>>
>>>>>>>On August 31, 2002 at 20:48:26, Vincent Diepeveen wrote:
>>>>>>>
>>>>>>>>On August 30, 2002 at 11:55:38, Robert Hyatt wrote:
>>>>>>>>
>>>>>>>>>On August 29, 2002 at 22:25:50, Mike Byrne wrote:
>>>>>>>>>
>>>>>>>>>>The much discussed game
>>>>>>>>>>
>>>>>>>>>>Deep Blue -- Fritz
>>>>>>>>>>8. World Computer Chess Championships (5) Chinese University HKG
>>>>>>>>>>1995.05.29 0-1 B33w
>>>>>>>>>>
>>>>>>>>>>1.e4 c5 2.Nf3 Nc6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 e5 6.Ndb5 d6 7.Bg5 a6 8.Na3 b5
>>>>>>>>>>9.Bxf6 gxf6 10.Nd5 f5 11.Bd3 Be6 12.Qh5 f4
>>>>>>>>
>>>>>>>>of course communication problem or not. doesn't take away
>>>>>>>>that castling is the losing move here. g3 is much much better.
>>>>>>>>another blunder caused by communication problems is not relevant
>>>>>>>>in a lost position. the castling lost the game already simply.
>>>>>>>>
>>>>>>>>beta DIEP needs less than a minute to get rid of o-o?? despite
>>>>>>>>running on background:
>>>>>>>
>>>>>>>That was the point. If you read Hsu's post about this years ago, he
>>>>>>>pointed out that Deep Thought had already found g3. But then the comm
>>>>>>>line went down before it moved, and they had to reconnect and re-start.
>>>>>>>It moved too quickly after losing so much time, and played right into
>>>>>>>the attack...
>>>>>>>
>>>>>>>It was unfortunate luck. But it also happened to me more than once. At
>>>>>>>the ACM event in 1978, were we played a horrible move that let Belle play
>>>>>>>a stunning sacrifice Rxh2. Against belle again in 1981 with the Qxb6
>>>>>>>loser...
>>>>>>>
>>>>>>>Thinking you are short on time can cause lots of problems..
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>>
>>>>>>>>00:00 262 0k 0 0 21 (2) 1 (0,0) 0.939 Na3xb5 a6xb5 Bd3xb5
>>>>>>>>00:00 454 0k 0 0 50 (2) 1 (0,0) 0.998 O-O-O
>>>>>>>>00:00 800 1k 0 0 120 (2) 2 (0,5) -0.498 O-O-O Nc6-d4
>>>>>>>>++ a3-b5
>>>>>>>>00:00 878 1k 0 0 202 (2) 2 (0,7) 0.939 Na3xb5 a6xb5 Bd3xb5
>>>>>>>>00:00 3471 3k 0 0 972 (2) 3 (0,55) 0.939 Na3xb5 a6xb5 Bd3xb5
>>>>>>>>00:00 10582 11k 0 0 3598 (2) 4 (1,106) -1.500 Na3xb5 a6xb5 O-O Nc6-d4
>>>>>>>>++ e1-c1
>>>>>>>>00:00 13372 13k 0 0 5884 (2) 4 (3,116) 0.343 O-O-O Be6xd5 e4xd5 Nc6-b4 Bd3xh7
>>>>>>>>Nb4xa2 Kc1-d2
>>>>>>>>++ e1-g1
>>>>>>>>00:00 14973 15k 0 0 8385 (2) 4 (4,130) 0.498 O-O Be6xd5 e4xd5 Nc6-b4
>>>>>>>>00:00 24761 25k 0 0 16095 (2) 5 (6,193) 0.948 O-O Be6xd5 e4xd5 Nc6-b4 Bd3-e4
>>>>>>>>00:00 43098 43k 0 0 40944 (2) 6 (12,303) 0.817 O-O Rh8-g8 Qh5xh7 Rg8-g7 Qh7-h5
>>>>>>>> Nc6-d4
>>>>>>>>00:03 71058 71k 0 0 230940 (2) 7 (39,551) 0.755 O-O h7-h6 c2-c3 Rh8-g8 Na3xb5
>>>>>>>>a6xb5 Bd3xb5
>>>>>>>>00:09 90278 90k 0 0 817017 (2) 8 (76,990) 0.142 O-O h7-h6 h2-h3 Be6xd5 e4xd5 N
>>>>>>>>c6-b4 Bd3-e4 Rh8-g8
>>>>>>>>00:25 115008 115k 0 0 2946515 (2) 9 (148,1659) 0.206 O-O h7-h6 h2-h3 Rh8-g8 c2-c
>>>>>>>>3 Rg8-g5 Qh5-f3 Be6xd5 e4xd5
>>>>>>>>00:52 126908 127k 0 0 6649989 (2) 10 (325,3074) 0.275 O-O h7-h6 c2-c3 Rh8-g8 h2-
>>>>>>>>h3 Be6xd5 e4xd5 Nc6-e7 Bd3-e4 Rg8-g5 Qh5-e2 f7-f5
>>>>>>>>++ g2-g3
>>>>>>>>01:25 135712 136k 0 0 11550480 (2) 10 (445,3720) 0.317 g2-g3 h7-h6 g3xf4 Rh8-g8
>>>>>>>>Bd3-e2 e5xf4 O-O-O Nc6-e5 Nd5xf4 Be6xa2 b2-b3
>>>>>>>>02:17 137275 137k 0 0 18937131 (2) 11 (741,5635) 0.511 g2-g3 h7-h6 c2-c3 Rh8-g8
>>>>>>>>Qh5-e2 Be6-g4 f2-f3 Bg4-e6 O-O-O Be6xd5 e4xd5
>>>>>>>>
>>>>>>>>>>13.O-O?? Rg8 14.Kh1 Rg6 15.Qd1 Rc8
>>>>>>>>>>{
>>>>>>>>>>[D] 2rqkb2/5p1p/p1npb1r1/1p1Np3/4Pp2/N2B4/PPP2PPP/R2Q1R1K w - -
>>>>>>>>>>
>>>>>>>>>>Although the game may already be lost with the infamous 13. O-O? , I also find
>>>>>>>>>>16. c4 ? to be a curiosity - is there any program that plays 16. c4 which allows
>>>>>>>>>>16. ...Qh4!.
>>>>>>>>>>}
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>16.c4 Qh4 17.g3 Qh3 18.Qd2 f3 19.Rg1 Rh6 20.Qxh6 Qxh6 21.cxb5 Bxd5 22.exd5 Nb4
>>>>>>>>>>23.Bf5 Rc5 24.bxa6 Nxa6 25.Nc2 Qd2 26.Ne1 Rxd5 27.Nxf3 Qxf2 28.Be4 Ra5 29.Rg2
>>>>>>>>>>Qe3 30.Re1 Qh6 31.Bc6+ Kd8 32.a3 f5 33.Rc2 Rc5 34.Rxc5 Nxc5 35.Rf1 Be7 36.a4 f4
>>>>>>>>>>37.gxf4 Qxf4 38.Rg1 Nxa4 39.b4 Qxb4 0-1
>>>>>>>>>>
>>>>>>>>>>Crafty 17.16 SE analysis after 16. c4
>>>>>>>>>>
>>>>>>>>>>White(1): 2rqkb2/5p1p/p1npb1r1/1p1Np3/2P1Pp2/N2B4/PP3PPP/R2Q1R1K b - c3
>>>>>>>>>>Black(1): go
>>>>>>>>>> clearing hash tables
>>>>>>>>>> time surplus 0.00 time limit 100:00 (100:00)
>>>>>>>>>> nss depth time score variation (1)
>>>>>>>>>> 1-> 0.00 -0.20 1. ... Qg5 2. Qf3 Bxd5 3. exd5
>>>>>>>>>> 2 0.00 -0.20 1. ... Qg5 2. Qf3 Bxd5 3. exd5
>>>>>>>>>> 2-> 0.01 -0.20 1. ... Qg5 2. Qf3 Bxd5 3. exd5
>>>>>>>>>> 3 0.04 0.01 1. ... Qg5 2. Rg1 b4 3. Nc2 Bxd5 4.
>>>>>>>>>> cxd5
>>>>>>>>>> 3-> 0.06 0.01 1. ... Qg5 2. Rg1 b4 3. Nc2 Bxd5 4.
>>>>>>>>>> cxd5
>>>>>>>>>> 4 0.06 0.01 1. ... Qg5 2. Rg1 b4 3. Nc2 Bxd5 4.
>>>>>>>>>> cxd5
>>>>>>>>>> 4-> 0.09 0.01 1. ... Qg5 2. Rg1 b4 3. Nc2 Bxd5 4.
>>>>>>>>>> cxd5
>>>>>>>>>> 5 0.17 -0.13 1. ... Qg5 2. Rg1 Rh6 3. g3 Bxd5 4.
>>>>>>>>>> exd5 bxc4 5. Nxc4
>>>>>>>>>> 5-> 0.21 -0.13 1. ... Qg5 2. Rg1 Rh6 3. g3 Bxd5 4.
>>>>>>>>>> exd5 bxc4 5. Nxc4
>>>>>>>>>> 6 0.67 -0.14 1. ... Qg5 2. g3 fxg3 3. fxg3 Bg4 4.
>>>>>>>>>> Be2 Bh3 5. Rf2 bxc4 6. Nxc4
>>>>>>>>>> 6-> 0.76 -0.14 1. ... Qg5 2. g3 fxg3 3. fxg3 Bg4 4.
>>>>>>>>>> Be2 Bh3 5. Rf2 bxc4 6. Nxc4
>>>>>>>>>> 7 1.39 -0.05 1. ... Qg5 2. g3 fxg3 3. fxg3 Bg4 4.
>>>>>>>>>> Be2 Bxe2 5. Qxe2 Nd4
>>>>>>>>>> 7-> 1.57 -0.05 1. ... Qg5 2. g3 fxg3 3. fxg3 Bg4 4.
>>>>>>>>>> Be2 Bxe2 5. Qxe2 Nd4
>>>>>>>>>> 8 2.70 -0.14 1. ... Qg5 2. g3 fxg3 3. fxg3 Bg4 4.
>>>>>>>>>> Be2 Bh3 5. Rf2 bxc4 6. Nxc4
>>>>>>>>>> 8 6.79 ++ 1. ... Qh4!!
>>>>>>>>>> 8 11.89 1.55 1. ... Qh4 2. g3 Qh3 3. Qf3 Rh6 4.
>>>>>>>>>> Qg2 Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1
>>>>>>>>>> 7. Rxf1
>>>>>>>>>> 8-> 11.89 1.55 1. ... Qh4 2. g3 Qh3 3. Qf3 Rh6 4.
>>>>>>>>>> Qg2 Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1
>>>>>>>>>> 7. Rxf1
>>>>>>>>>> 9 15.00 1.54 1. ... Qh4 2. g3 Qh3 3. Qf3 Rh6 4.
>>>>>>>>>> Qg2 Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1
>>>>>>>>>> 7. Rxf1 b4
>>>>>>>>>> 9-> 15.51 1.54 1. ... Qh4 2. g3 Qh3 3. Qf3 Rh6 4.
>>>>>>>>>> Qg2 Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1
>>>>>>>>>> 7. Rxf1 b4
>>>>>>>>>> 10 21.87 1.49 1. ... Qh4 2. g3 Qh3 3. Qf3 Rh6 4.
>>>>>>>>>> Qg2 Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1
>>>>>>>>>> 7. Rxf1 fxg3 8. hxg3
>>>>>>>>>> 10-> 24.57 1.49 1. ... Qh4 2. g3 Qh3 3. Qf3 Rh6 4.
>>>>>>>>>> Qg2 Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1
>>>>>>>>>> 7. Rxf1 fxg3 8. hxg3
>>>>>>>>>> 11 43.46 1.61 1. ... Qh4 2. g3 Qh3 3. Qf3 Bg4 4.
>>>>>>>>>> Qg2 Qh5 5. f3 Bh3 6. Qd2 Bxf1 7. Rxf1
>>>>>>>>>> fxg3 8. cxb5 axb5 9. Nxb5 gxh2 10.
>>>>>>>>>> Qxh2
>>>>>>>>>> 11-> 47.00 1.61 1. ... Qh4 2. g3 Qh3 3. Qf3 Bg4 4.
>>>>>>>>>> Qg2 Qh5 5. f3 Bh3 6. Qd2 Bxf1 7. Rxf1
>>>>>>>>>> fxg3 8. cxb5 axb5 9. Nxb5 gxh2 10.
>>>>>>>>>> Qxh2
>>>>>>>>>> 12 1:32 1.89 1. ... Qh4 2. g3 Qh3 3. Qf3 Bg4 4.
>>>>>>>>>> Qg2 Qh5 5. f3 Bh3 6. Qe2 Bxf1 7. Rxf1
>>>>>>>>>> Rxg3 8. cxb5
>>>>>>>>>> (2) 12-> 2:28 1.89 1. ... Qh4 2. g3 Qh3 3. Qf3 Bg4 4.
>>>>>>>>>> Qg2 Qh5 5. f3 Bh3 6. Qe2 Bxf1 7. Rxf1
>>>>>>>>>> Rxg3 8. cxb5
>>>>>>>>>>
>>>>>>>>>>==========================================================================
>>>>>>>>>> Crafty SE 17.16 analysis after 15. ...Rc8 16.f3 or 16. g3 seem to offer
>>>>>>>>>>stronger resistance
>>>>>>>>>>
>>>>>>>>>>[D] 2rqkb2/5p1p/p1npb1r1/1p1Np3/4Pp2/N2B4/PPP2PPP/R2Q1R1K w - -
>>>>>>>>>>
>>>>>>>>>>White(1): go
>>>>>>>>>> clearing hash tables
>>>>>>>>>> time surplus 0.00 time limit 100:00 (100:00)
>>>>>>>>>> nss depth time score variation (1)
>>>>>>>>>> 2 0.00 -1.32 1. Kg1 Bh3 2. g3 Bxf1 3. Bxf1 fxg3
>>>>>>>>>> 4. hxg3
>>>>>>>>>> 2 0.00 ++ 1. Be2!!
>>>>>>>>>> 2 0.00 0.19 1. Be2 Qg5 2. Bf3 Bxd5 3. Qxd5
>>>>>>>>>> 2-> 0.02 0.19 1. Be2 Qg5 2. Bf3 Bxd5 3. Qxd5
>>>>>>>>>> 3 0.02 0.08 1. Be2 Qg5 2. Bf3 Bg7
>>>>>>>>>> 3 0.03 0.11 1. Qh5 b4 2. Nc4 Bxd5 3. exd5
>>>>>>>>>> (2) 3-> 0.05 0.15 1. Re1 Qg5 2. Qf3 Bg4
>>>>>>>>>> 4 0.06 0.04 1. Re1 Qg5 2. Bf1 Bg4 3. Qd3
>>>>>>>>>> 4 0.08 0.09 1. Be2 Qg5 2. Rg1 Nd4 3. Bh5
>>>>>>>>>> (2) 4-> 0.10 0.09 1. Be2 Qg5 2. Rg1 Nd4 3. Bh5
>>>>>>>>>> 5 0.11 0.02 1. Be2 Qg5 2. Rg1 Bxd5 3. Qxd5 Bg7
>>>>>>>>>> (4) 5-> 0.24 0.02 1. Be2 Qg5 2. Rg1 Bxd5 3. Qxd5 Bg7
>>>>>>>>>> (3) 6 0.30 -0.19 1. Be2 Qg5 2. Bf3 Nd4 3. Re1 Bxd5 4.
>>>>>>>>>> exd5
>>>>>>>>>> (2) 6 0.49 -0.12 1. c4 Qg5 2. Rg1 Qh4 3. Rf1 Rh6 4.
>>>>>>>>>> h3 Bxd5 5. exd5
>>>>>>>>>> 6-> 0.78 -0.12 1. c4 Qg5 2. Rg1 Qh4 3. Rf1 Rh6 4.
>>>>>>>>>> h3 Bxd5 5. exd5
>>>>>>>>>> 7 0.97 -0.07 1. c4 Qg5 2. Rg1 Qh4 3. Rf1 b4 4. g3
>>>>>>>>>> fxg3 5. fxg3
>>>>>>>>>> (3) 7-> 1.67 -0.07 1. c4 Qg5 2. Rg1 Qh4 3. Rf1 b4 4. g3
>>>>>>>>>> fxg3 5. fxg3
>>>>>>>>>> 8 4.41 -- 1. c4
>>>>>>>>>> (2) 8 5.27 -0.69 1. c4 Qh4 2. cxb5 Rh6 3. Nf6+ Kd8 4.
>>>>>>>>>> h3 f3 5. bxc6 fxg2+ 6. Kxg2 Bxh3+ 7.
>>>>>>>>>> Kg1 Bxf1
>>>>>>>>>> 8 9.19 -0.30 1. Qd2 Qh4 2. f3 Bg7 3. Kg1 Kf8 4.
>>>>>>>>>> Rae1 Kg8 5. Be2
>>>>>>>>>> 8 11.31 -0.21 1. Re1 Qh4 2. Kg1 Nd4 3. c3 Qh3 4.
>>>>>>>>>> g3 Bxd5 5. exd5
>>>>>>>>>> (3) 8-> 12.94 -0.21 1. Re1 Qh4 2. Kg1 Nd4 3. c3 Qh3 4.
>>>>>>>>>> g3 Bxd5 5. exd5
>>>>>>>>>> 9 13.92 -- 1. Re1
>>>>>>>>>> (2) 9 16.77 -1.24 1. Re1 Qh4 2. Kg1 Nd4 3. Be2 Bh3 4.
>>>>>>>>>> Bf1 f3 5. Ne3 Qxe4
>>>>>>>>>> 9 17.13 ++ 1. Qd2!!
>>>>>>>>>> 9 37.08 -0.35 1. g3 fxg3 2. fxg3 Nd4 3. c3 Bg4 4.
>>>>>>>>>> Qd2 Bf3+ 5. Kg1 Bh6 6. Qf2
>>>>>>>>>> 9 47.50 -0.31 1. f3 Qh4 2. Qe1 Qg5 3. Qf2 Be7 4.
>>>>>>>>>> Rae1 Nd4 5. Kg1 Bh3
>>>>>>>>>> (2) 9-> 50.02 -0.31 1. f3 Qh4 2. Qe1 Qg5 3. Qf2 Be7 4.
>>>>>>>>>> Rae1 Nd4 5. Kg1 Bh3
>>>>>>>>>> 10 1:08 -0.26 1. f3 Qg5 2. Qd2 Rh6 3. Kg1 Qh4 4.
>>>>>>>>>> h3 Rg6 5. Nb6 Rb8 6. Nd5 Bxd5 7. exd5
>>>>>>>>>> 10 1:15 2/36* 1. g3
>>>>>>>>>>
>>>>>>>>>>===========================================================================
>>>>>>>>>>
>>>>>>>>>>Force analysis on 16. c4?
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>White(1): search c4
>>>>>>>>>>White(1): go
>>>>>>>>>> time surplus 0.00 time limit 100:00 (100:00)
>>>>>>>>>> nss depth time score variation (1)
>>>>>>>>>> 5 0.00 -0.32 1. c4 Qg5 2. g3 fxg3 <HT>
>>>>>>>>>> 5-> 0.00 -0.32 1. c4 Qg5 2. g3 fxg3 <HT>
>>>>>>>>>> 6 0.00 -0.32 1. c4 Qg5 2. g3 fxg3 <HT>
>>>>>>>>>> 6-> 0.00 -0.32 1. c4 Qg5 2. g3 fxg3 <HT>
>>>>>>>>>> 7 0.02 -0.32 1. c4 Qg5 2. g3 fxg3 <HT>
>>>>>>>>>> 7-> 0.02 -0.32 1. c4 Qg5 2. g3 fxg3 <HT>
>>>>>>>>>> 8 0.95 -- 1. c4
>>>>>>>>>> 8 1.05 -0.69 1. c4 Qh4 2. cxb5 Rh6 3. Nf6+ Kd8 4.
>>>>>>>>>> h3 f3 5. bxc6 fxg2+ 6. Kxg2 Bxh3+ 7.
>>>>>>>>>> Kg1 Bxf1
>>>>>>>>>> 8-> 1.05 -0.69 1. c4 Qh4 2. cxb5 Rh6 3. Nf6+ Kd8 4.
>>>>>>>>>> h3 f3 5. bxc6 fxg2+ 6. Kxg2 Bxh3+ 7.
>>>>>>>>>> Kg1 Bxf1
>>>>>>>>>> 9 3.03 -- 1. c4
>>>>>>>>>> 9 10.80 -1.55 1. c4 Qh4 2. g3 Qh3 3. Qf3 Rh6 4. Qg2
>>>>>>>>>> Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1 7. Rxf1
>>>>>>>>>> 9-> 10.80 -1.55 1. c4 Qh4 2. g3 Qh3 3. Qf3 Rh6 4. Qg2
>>>>>>>>>> Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1 7. Rxf1
>>>>>>>>>> 10 14.49 -1.54 1. c4 Qh4 2. g3 Qh3 3. Qf3 Rh6 4. Qg2
>>>>>>>>>> Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1 7. Rxf1
>>>>>>>>>> b4
>>>>>>>>>> 10-> 14.49 -1.54 1. c4 Qh4 2. g3 Qh3 3. Qf3 Rh6 4. Qg2
>>>>>>>>>> Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1 7. Rxf1
>>>>>>>>>> b4
>>>>>>>>>> 11 23.28 -1.49 1. c4 Qh4 2. g3 Qh3 3. Qf3 Rh6 4. Qg2
>>>>>>>>>> Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1 7. Rxf1
>>>>>>>>>> fxg3 8. hxg3
>>>>>>>>>> 11-> 23.28 -1.49 1. c4 Qh4 2. g3 Qh3 3. Qf3 Rh6 4. Qg2
>>>>>>>>>> Qxg2+ 5. Kxg2 Bh3+ 6. Kg1 Bxf1 7. Rxf1
>>>>>>>>>> fxg3 8. hxg3
>>>>>>>>>> 12 39.97 -- 1. c4
>>>>>>>>>> 12 2:46 -2.00 1. c4 Qh4 2. g3 Qh3 3. Qf3 Bg4 4. Qg2
>>>>>>>>>> Qh5 5. Nb6 Bf3 6. Nxc8 fxg3 7. fxg3
>>>>>>>>>> Bxg2+ 8. Kxg2
>>>>>>>>>> 12-> 2:46 -2.00 1. c4 Qh4 2. g3 Qh3 3. Qf3 Bg4 4. Qg2
>>>>>>>>>> Qh5 5. Nb6 Bf3 6. Nxc8 fxg3 7. fxg3
>>>>>>>>>> Bxg2+ 8. Kxg2
>>>>>>>>>>
>>>>>>>>>>===========================================================================
>>>>>>>>>>
>>>>>>>>>>so after only 2 minutes on a micro search, maybe 1M nps, Crafty SE 17.16 sees
>>>>>>>>>>16. c4 is not looking so well ...the DB prototype. seeing 3-5M nps did not see
>>>>>>>>>>this coming?
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>The problem, as is well-known, was that they had a communication crash during
>>>>>>>>>this game. And once they noticed and got everything set back up, Deep Thought
>>>>>>>>>was left with little time on the clock to do the search.
>>>>>>>>>
>>>>>>>>>I've mentioned more than a couple of times this happened to me in ACM events...
>>>>>>>>>
>>>>>>>>>It comes with the territory of accessing a machine remotely, on a brand new
>>>>>>>>>network (was just set up in Hong Kong that week and had significant but unknown
>>>>>>>>>problems).
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.