Author: Robert Hyatt
Date: 20:30:29 07/29/03
Go up one level in this thread
On July 29, 2003 at 16:25:38, Vincent Diepeveen wrote: >On July 29, 2003 at 00:54:39, Keith Evans wrote: > >>On July 29, 2003 at 00:31:17, Robert Hyatt wrote: >> >>>On July 28, 2003 at 20:59:24, Vincent Diepeveen wrote: >>>>It is like comparing a sniper rifle from 2003 with a sniper rifle from world war >>>>1. >>>> >>>>Distances they shot at in world war 1 and 2 with sniper rifles must have been a >>>>few hundreds of meters. >>> >>>In WW1 my grandfather was a sniper. He shot at ranges up to 1000 yards. >>> >>>In WW2 my father was a sniper. He shot at ranges up to 1000 yards. >>> >>>Today, a neighbor down the street is a sniper. He shoots at ranges up to 1000 >>>yards. >>> >>>_nobody_ shoots a sniper rifle at ranges of "kilometers" today. "kilometer" >>>perhaps. With an occasional attempt at up to 2km with a big 50 cal "rifle". >>> >> >>Supposedly Gunnery Sergeant Hathcock took out an NVA at 2500 yards with a .50 >>caliber machine gun. A friend got into the whole "Marine Sniper" scene and it >>was a little scary. Nice skill to have if you need it, but it scares me when >>people fantasize about it. (Especially when the word "safety" is spelled >>incorrectly at the range ;-) >> >>>This is just another area where you know nothing, but write as though you are >>>an expert. >>> >>>BTW, Hsu's move generator is _not_ a lot better than Belle. All you have to >>>do is read his paper to see what he did... > >>Hsu did add some modes which Belle did not have. For example finding checking, >>check evasion, and attacking moves. He hints at some other things like >>generating moves for pruning but this is very vague. He may have handled those > >It is not vague. Read his writings in artificial intelligence. He's doing no >progress pruning in hardware. Also it seems to forward prune last few plies in >hardware too. In fact it is very clever to forward prune in hardware, even >though it must be rude code. > >Note that this forward pruning i refer to (so not the progress pruning) is not >doing in move generator. > >>basic special case moves (castling, ep captures) more elegantly, but it's hard >>to tell without seeing the implementation details of each. Hsu also added >>hardware repetition detection which is not part of the basic move generator >>logic, but if you group it with movegen just for the sake of argument then it's >>a noteworthy improvement. Maybe the adjective that Vincent used was a little > >Look we can discuss long, but without repetition detection you are at the >absolute beginners level from software viewpoint. > >Especially when doing checks in qsearch doing repetition detection is very >important. > >It is a sequential clock though for the search. So it slows you down quite >something. > >>extreme, but this statement doesn't bother me too much. > >People do as if deep blue is a holy thing. Which ain't true. It's gnuchess in >hardware. Now in 1997 that was of course a magnificent achievement. Compliments >to Hsu for that. I have said it a lot. > >Imagine how hard his job was. And getting it a bit parallel to work. AND in >hardware at tens of millions of nodes a second, which is what the sponsor >wanted. > >Also we have the statement of Hyatt that the old deep blue 1 was a very poor >piece square table program. Can we get this right once and for all? the _first_ chiptest chip was a piece/square evaluation. The next deep thought chip was more complex, although it did use piece/square tables. Deep Blue was another re-design with much more in the eval besides simple piece/square. Deep Blue II was _another_ re-design. Stop saying deep blue when you mean chiptest/deep thought. > >So going from a very poor piece square table program to something that is at the >gnuchess level is really a *major* step forward. > >Note that to get 10-11 ply the 1998 versions of zarkov (of course the improved >gnuchess version, as John Stanback privately worked on at gnuchess) needed like >1 hour to 2 hours a move. > >So that just shows how good the job from Hsu was. > >Note that Zarkov was at the time one of the stronger programs on the planet. So >putting gnuchess evaluation and ideas in hardware was not at all a bad idea! > >The qsearch and the extensions were done way smarter though, which means that >Deep Blue tactical was a lot better than that. > >But it is wrong to suggest that it still would be a formidable opponent. > >take 20 times faster hardware than a 200Mhz Pro and take a newer zarkov version >which is using a bit less time now to get to that depth. > >Because the reason why zarkov had problems getting to 10-12 ply in those days is >the same reason why deep blue just got 10-12 ply. It was NOT using nullmove a >lot! > >Zarkov did have it, but he did some fullwidth nonsense last few plies. > >I do not know whether he has improved it sincethen. I guess so. > >It is trivial that going from deep blue 1 (piece square table program) to deep >blue 2 was a major step forwards. > >It is also trivial that comparing with 2003 software and hardware it looks >completely outdated. > >Hsu never was a chessprogrammer. He was a hardware guy. Have you _ever_ talked to him? I have. He was not "just a hardware guy." He came up with the idea of SE before _anybody_ was doing it. That's not "just hardware". >So if a commercial programmer like Donninger is working for a year or 4 at a >hardware version of nimzo now, then it is trivial that this is going to look a >lot better than the oldies looked like. > >>-K
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.