Author: Uri Blass
Date: 06:48:34 04/04/04
Go up one level in this thread
On April 04, 2004 at 09:00:45, Vincent Diepeveen wrote: >On April 03, 2004 at 01:48:15, Jonas Bylund wrote: > >>>You mean whether it's textmode or graphical i guess. >>> >>>It's graphical GUI :) >> >>So i guess it runs in DOS mode ;) >>Will it run under Winboard? (i know i have asked this before, but i use WinXP >>exclusively and i really don't like DOS) > >You mean whether i can load winboard or winboard engines into my interface? > >> >>>Success is relative. I do not earn a penny on EGTBs. I just waste time in fact >>>on it. I could earn money in that time instead. So success is relative. > >>Indeed success is relative and in this case it is relative to the existing >>format, compared to that, 10-20 Gb would be hugely successful. However if you > >I do not see why, all you need is a better compression. That requires immense >amounts of cpu time though. > >Call Nalimov a pioneer, or call him one-eyed king in the land of the blinds. > >Probably both is true. > >Same for Thompson and others. > >If just 1 person has been busy, it's easy to improve upon him. > >Look Eugene has never been busy with efficient storing his EGTBs. I'm sure he >can come up with efficient storing too if he wanted to. > >However he'll need to write pretty efficient software to even at a few dual and >quad machines to get it compressed well. > >Compression is more complex than computerchess in itself. > >>mean success relative to your standards it is a different story of course, and i >>mean that in a general sense. >> >>If you measure success relative to money, then IMO i would say you have failed >>already, since true reward does not need reckognition nor payment. > >Relative to success, all computerchess guys are major failures. > >Selling second hand cars earns a factor X, X > 100, more money. > >>>I do not see it as a success. Getting even 5th in world champs is way harder. >>>Getting #300 on ATP list in tennis even harder than that. I have in total >>>written 3 different generators now for egtb's and each project was only a few >>>days worth. Very easy to make in fact. I'm sure any programmer who has past >>>beginners stage can make it. >>> >>>If you compare that with a chess engine, a chessengine is 1000x harder to get to >>>a strong level. >>> >>>Even generating the first 7 men which i expect to do real soon i do not consider >>>as a big success. I'm sure Eugene could do that too knowing his huge hardware. >>>However i plan to make a special generator for the 7 man a lot faster than any >>>generator before. I had a while ago a generator which was very fast. >>> >>>It's easier than some guess to do all that. The problem is it is all volunteer >>>work in some sense. Also the importance for chessprograms of egtb's is smaller >>>nowadays than it used to be say in 1999. >>> >>>The impact of the first programs using 5 men was way bigger than 6 men now. >>> >>>The reason is the huge improvement of chess software (the top 10 of the world >>>that is). >>> >>>I remember junior playing at world champs 2003 with just all 5 men turned on. >>> >>>Note i didn't use any egtb at the world champs 2003. > >>That's interesting, why not? especially when you have such compressed tb's. > >I had bigger priorities at the supercomputer to say it polite. Like praying that >it directly would run smooth. > >Secondly a supercomputer has a limited number of file handles. > >So 1000 files x 512 processors = 512k file handles. > >Forget it, no supercomputer has 512k file handles. > >In diep i open the file handles before the program starts. > >Nalimov same thing. > >I just used uncompressed EGTBs at supercomputer. Terabytes of diskspace! > >But starting it took 3 hours. Then i had 2 crashes. So i turned them off for the >rest of the tournament. No more crashes then. > >1 day later entire machine was crashed for half a day or a full day (i do not >know exactly). There was only 1 round that day. > >I'm 100% sure it was a result of the huge amount of allocated file handles and >SGI file system. Their XFS sucks a lot. I understand why they are moving to >linux. It's just a lot better! > >Only the 2.6.4 kernel is not so very good yet. It requires more work to get it >more stable. Yet we know it *will* happen. The sgi stuff will *not* improve at >that speed simply. > >The principal of having shared file handles, despite that i just want local file >handles, it sucks if i may say so. > >When i say shared file handles i mean that there is a few central processors >serving the other 500 processors for files. There is 1 clock processor, and so >on. > >In short, to do i/o which can be done in great ways in such machines, one needs >an entire different approach than what is ideal for computerchess at a PC. > >I'm just not willing to rewrite my routines just for 1 championship and i had >decided that back in start 2003 already. > >>>In both cases i do not see the problem. >>> >>>Also note that where i love small egtb's, chessbase shows up with huge machines >>>with a lot of ram and harddisk packs. >> >>I believe it was Lasker who once said "Play the board not the man" i find new >>uses for that phrase on a daily basis, in this case i would say that all the >>time and energy that are potentially wasted on focusing on your opponent/s, >>could be invested in your own development and preparation. > >EGTBs are in that sense a waste of time yes. That's why most just copy the >one-eyed routines and use them and praise its programmer. > >>>I just cannot afford harddisk packs as i have no sponsor. I'm sure they keep on >>>using nalimov for some years to go. No need to change. The effect of the egtb's >>>is getting less anyway. The current generation of software is not stupid enough >>>to allow to get to an endgame which is dead lost. >>> >>>Even when some hard egtb's come there: >>> >>>for example what happens sometimes is: >>> >>> KRP KRP >>> >>>it is very *unlikely* that by accident some commercial program goes to an >>>endgame which is lost. if it is forced in such an endgame it probably is already >>>down more than minus 2, so it not exactly volunteerly went into that position. >> >>Sometimes winning a +2 endgame can be really hard, even for computers and i >>would be very surprised if the programX with 3+4+5+6+7 men tb's wouldn't win >>that scenario more frequently than programY without the tb's > >You are correct for programs released in 1997. > >Yes even 1999. > >But sincethen they have improved sufficiently. > >The whole point is that there must be an alternative to draw it. > >If that is not there, then having egtb's doesn't matter. > >If it is there, then they matter. > >Your guess for the future is as good as mine. > >But the real point is that EGTBs can help to save a lost game. > >Nevertheless, it's better not having that lost position. So all efforts should >be in opening and middlegame and start of endgame more than far endgame. > >If you have a few pieces and a few pawns, then accessing EGTBs is just a waste >of time. > >Only when the game is already decided objectively, then EGTBs jump in and can >help you save a lost position. > >So from absolute viewpoint seen, programs like junior have a playstyle which >will hardly get them to positions where having EGTBs matter. > >EGTBs just matter when the program plays like a beginner, do not forget that. > >>>the evaluations have advanced too much to let them get fooled easily. i noticed >>>that fritz is doing a good job there especially. it has improved a lot in rook >>>endgames the last so many versions. i was a live witness of that at world champs >>>2002 in fact. diep managed to get some won rook endgame (rpp versus rp) but >>>fritz very convincingly drew it. diep had no clue it was a draw. fritz did. >> >>Well i guess it wasn't a won endgame then :) > >it was won somewhere when both were nullmoving. > >But the position in question would have needed 7 men at least to win it. > >KRPP KRP to be precise. > >>>Even for yace i doubt whether any rook endgame EGTB will save its ass somewhere. >>>It usually is doing it itself much better :) >>> >>>>I am not implying that we all have a similar standard when it comes to giving >>>>our work away, but people who know me fairly well, knows that if they need >>>>anything, that be music, webdesign, painting, ideas etc.all they need to do is >>>>ask and i will provide it for free if i have the time. Not too long ago someone >>>>saw one of my paintings (he lives in USA) and really liked one particular >>>>painting and i sent it to him free of charge, - shipping charges. >>>>I can say that giving without asking anything in return is _extremely_ >>>>liberating, but if that dosn't do it for you, i respect that too. >>> >>>Why would i provide to competitors for free source code of egtb's? > >>Because IMO the more "things" we hang on to in our lives, the more tied up we >>get, the happiest people i know are people who give/share without strings or >>conditions. That was the philosophical answer, now on to the practical; (which > >Why not email to chessbase for their interface source code, so you can fix their >UCI support. > >>might end up a bit philisophical too) if you where to give away the tb's and the >>source for people to freely impliment support and download, then your >>competitors would be stronger than before, allthough it seems you think it won't >>make a difference, and if they indeed would be stronger with your 3-7 men tb's >>then you would have to work harder to make Diep stronger, that competition could >>benefit Diep in the long run right? > >I didn't read a single sentence from between where i wrote 'uci' and here. > >>>Note i wouldn't mind a few amateur programs, but what happens is the format >>>spreads, that's what happens. >>> >>>Then competitors use it and they do not say 'thank you'. They just use it. >> >>Even if that was true, then so what? the analogy "If a tree falls and no one >>hears it, did it make a sound?" (not sure i got it exactly right to the letter, >>but the essence should be pretty clear) to me the meaning is multidimensional, >>but it asks a very interesting question which with a bit of interpeting looks >>like this: "in order to feel/be rewarded and satisfied we need someone to >>record/reward our acomplishment?" a goal of mine in my life is to reach a point >>where i can "chop that tree" and be satisfied/rewarded by the task itself >>regardless if anyone records it. > >>>They have always superior hardware and huge harddisks. >> >>LOL i remember 500 processors being used for, eeeh i forgot the name of the >>engine... no i remember it was Diep :) > >Testing is the most important thing in computerchess i guess. I've never done >that lack the personnel to do it in fact. And some article writers are not much >of a help either there. Testing at a 500 processor machine forget it. You complained about bugs in the evaluation. Could not you use the same evaluation in a slower machine and test it? > >For the same reason why deep blue played like such a joke. It played better than the opponents at that time(I do not claim that it played better than the programs of today) > >In 1997 it would not have won any tournament. Too poor tested. I believe that it was stronger than other programs in 1997. > >In fact it already failed in 1995. In 1995 it was a different hardware. Uri
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.