Author: Uri Blass
Date: 00:40:21 08/26/01
Go up one level in this thread
On August 25, 2001 at 22:27:05, Vincent Diepeveen wrote: >On August 25, 2001 at 19:55:45, Dann Corbit wrote: > > >>You missed out on this one: >>Tao >> Single >> Pentium III 700 MHz > >>But you did play a 866 MHz machine: > >>[Event "18th WMCC"] >>[Site "6th Olympiad Maastricht"] >>[Date "2001.08.22"] >>[Round "7"] >>[White "Diep, 2x AMD 1.2"] >>[Black "XiniX, Intel 866"] >>[Result "1-0"] >>[ECO "E37"] >>[PlyCount "107"] >>[EventDate "2001.??.??"] >>[Source "Frank Quisinsky"] > >Ah at my notation paper it reads: 1Ghz K7 for Xinix, but >even with a 100 point rating increase it still would not have >profited much, as some bugs in engine were the problem here, not >the hardware. > > >>>Secondly, everyone who can make a chessprogram can make his program parallel >>>without much effort. >> >>How long did it take you? If someone has never done SMP programming it is not a >>trivial thing. >> >>>Third, some people, including stefan meyer kahlen, will find out that their >>>program is single cpu faster than dual. >> >>This is not surprising. I suspect it will take a year to get SMP extremely well >>implemented. It depends, of course, on how many global variables there are that >>are written to and how many static variables that are written to as well. > >You are living in the old days when search was split over several processors. > >Any algorithm that is going to split search is going to be hard to make, >however the algorithm as i posted here at CCC, which doesn't share anything >but hashtable and only needs somehow to transfer the move you did to the >other process, that's a very simple thing to implement. > >About half a day to get it working, and another day to bugfix. > >Even the code to share hashtable i can cut'n paste here if you want to, >linux and/or windows. > >>>If SMK was wrong here then it was a political decision to go for the >>>single title. >> >>It was obviously a good idea because it succeeded. > >from chessbase viewpoint sure, from SMK viewpoint i am not so sure. > >because if you walk into a shop there is a box saying: > 1. Junior 8 > 2. Fritz 6 > 3. Shredder 6 > >Sorry stefan that i played you and only scored 4 points otherwise your >buchholz would've been better! > >The only person dicked here is stefan. > >Because if he would have had a dual most likely he would at least have won >against > a) diep (the qxd8? bad move in opening by shredder against diep > was made too quick and a dual would've gotten > a fail low here for sure after which i end up with a bad > position with diep) > b) junior > >So that would've been already from head a 7.5 out of 9 score instead >of 6 out of 9 for Stefan. It is not so simple. I am not sure if it could win the games that you mentioned with 2 processors and even if it could win it does not mean that it could get at least 7.5 because winning means playing against better opponents that shredder did not play. Uri didn't look at his other points lost even, >also i think that a book loss is a book loss, no matter hardware. > >Bookloss was versus kure. > >>>Considering Junior won the tournament, i think that's enough proof that >>>it was not so smart to go single cpu from Stefan, but well his official >>>statement is that a 1.4Ghz machine K7 is faster for him than a dual 1Ghz >>>intel. >> >>Maybe he knew he had a much better chance to get an official title and not even >>have to worry about Junior or Fritz (or Ferret or Crafty or Diep). In fact, I >>think his choice was brilliant. Of course, there was Chess Tiger, but in the >>end, he will get to write: >>"Three times consecutive computer chess world champion" on his software box. >>Can anyone really believe that he made a mistake? > >Yes a major mistake. > >Shredder sales will be crunched by this major blow this year. Every idiot >can see that junior won this tournament with a huge score and that even >fritz ended higher than shredder (even though we know it's only sum >of opponents). > >>I think the multiple CPU championship was probably a LESS PRESTIGIOUS victory. > >No it's the most important victory as it wins the tournament. > >>I am talkiing about an advertizing standpoint. If I see that a program won a >>contest with more than one CPU and my machine only has one CPU, I might very >>well decide to try the product that wins with only one CPU. Since that is >>99.99% of the market, which segment would you target? >> >>All in all, I think SMK out-thought everyone except Christophe. > >I think he was very stupid, and he is stupid if he listens to chessbase >next year again. > >>>Some smarto's here at CCC can verify this easily. >>> >>>Note that i think that Stefan lost loads of points onto book. Like the >>>dubious line i played against it would never have even lost 0.5 point >>>from a strong book. > >>Points lost or no points lost, he still won the single CPU world championship. >>The bottom line is all that matters. > >>>He lost on book to fritz, and Stefans really only unlucky game i think >>>was against junior, dual he might already have at least drawn the game. >>> >>>I'm btw amazed that fritz is faster on a dual 1Ghz intel than dual 1.2Ghz AMD. >>>Perhaps speed and getting another ply doesn't matter that much anyway. > >>I have seen a similar effect for some programs. For instance, Eugene Nalimov's >>build of Crafty seems faster on a 900 MHz Intel box than on my 950 MHz AMD. >>The builds I make are not like that, but it seems to depend on a lot of factors. > >the new visual c++ sp5 + processor pack is hell faster on a K7 than on an >intel, i can't imagine that a C(++) program like shredder is faster on >an intel. > >Quest is completely made in assembly, which explains why fritz is faster >on a P3 probably. Also for shredder speed is definitely more interesting >than fritz, because of shredder's passive play which requires to outsearch >the opponent otherwise he's history. this where fritz might perhaps lack >so called 'well known' chess knowledge, but it plays a very active >way of chess. besides being good for your heart to play healthy chess, >it also means that it doesn't need to outsearch its opponent at all to >win. > >Note it seems to me that quest on a Pii isn't much faster than on a P3 when >the Mhz gets calculated back. For me the diff between katmai and coppermine >is about 17.3% though. > >>>A pc is something which is not too big and which you can carry. Duals >>>are very cheap. A single cpu Xeon is more expensive than a dual P3. >> >>I agree that duals are a very smart buy. But most people won't do it for some >>strange reason. The next computer I buy out of pocket will have more than one >>CPU (almost for sure). > >We completely agree here! > >In case you go to a shop and plan to buy a dual AMD, just get the >latest stepping tbirds instead of the expensive palomino's. palomino's >are faster, but latest steppings tbirds work great too! > >>>A dual AMD isn't that expensive either. I paid $3000 for my thing (1Gb >>>registered DDR ram) but soon way cheaper dual AMDs will be there (without >>>needing registered ecc-ddr ram and without needing probably a 460+ watts >>>power supply with 24 pins e-atx instead of the way cheaper >>>standard 20 pins atx). Note that this $3000 includes a LCD screen, >>>a server case (slightly bigger as a bigtower) with wheels and it's lighter >>>than my pentiumpro200 internet computer which weighs 40 kilo's. >>> >>>I'm amazed by the big protests against being dual. The only valid protest >>>is that the allowance of multiprocessors was allowed. This amazed me too, >>>because that would allow quads and octo processors. >> >>If a dual is a problem, then a faster CPU is a problem. After all, it's the >>same defect -- the one with the fastest machine gets free ELO. If you are 4 >>times faster, you just got +100 ELO at least. A 2400 IM verses a 2500 GM in a >>contest -- who will come out on top? You could think of it as making the slower >>programs play at knights odds. Ouch. In particular, XiniX was at a HUGE >>disadvantage. The bigger cache and more advanced chips of the faster machines >>makes the dropoff much larger than the MHz difference. > >Xinix would not have profited much from faster hardware i can tell you >that. Bugs in programs completely overrule speed. > >Xinix did get only 9 ply in openings, but it didn't do any forward pruning. > >He'll shout out loud now, but i know from experience that from 9 ply till 11 >ply a bit of bugfixing is definitely needed. > >Also i'll not complain that some 'crucial' moves i only got 10 ply, >like against gandalf. > >Bugs completely overwhelm search depth. A better eval is always better than >another 4 ply when you're at 10 ply + loads of extensions without forward >pruning other than nullmove. > >I also had in all games the impression that i was tactical way stronger >than my opponent. Except for the stupid exchange to endgame against Gandalf >i had the feeling i tactical/positional saw it all sooner than my opponent! > >Considering my search depth usual wasn't bigger than my opponent, the big >bulk of those scores must have been positional! > >>>Now by accident a dual AMD is way faster than any quad for me, but >>>for me a quad is definitely not in the same price league. It definitely >>>can be in the 'carryable' league. >> >>I find this very strange. Do you even mean a 900 MHz quad? That would be >>pretty bizarre. > >which 900Mhz quad? > >All the 900Mhz xeon processors are called back to factory because of >a major design flaw. > >fastest quad is a 700Mhz quad. > >But yes a 1.5Ghz palomino dual is for me faster than a quad 900 for sure. > >>>Anyway what i would like most is that all future tournaments from ICCA >>>are completely open hardware, the days that any supercomputer who joined >>>would get ahead of all micro computers are over anyway. > >>Will you still feel that way if 8 CPU Siemens 2GHz machines show up, giving an >>effective throughput of 8*2*.75 = 12 GHz, and you are still at 2GHz? They have >>a factor of 6 speed increase, which would be several hundred ELO. Imagine a 10 >>Gigabyte hash table that can be read in a flash. Imagine 7 men tablebase files >>on 15K RPM drives with lots of ram to cache them (maybe 100 GB ram for cacheing >>tablebase files). Will you compete effectively? Is it still a PC when it costs >>$100K? > >In fact i prefer to play zugzwang, zugzwang at a 500 processor 21264 600Mhz >21264 alpha. > >That's about 300Ghz or so. > >Because a world championship IMHO is about letting the best programs play, >no matter hardware. If a factor 4 speedup would be 100 rating points >according to your calculations, then what rating would >Zugzwang be at? > >Please consider that it lost from lambchop at wcc99. lambchop was searching >most of the moves 8 ply and ran at a 450Mhz PII. > >Zugzwang ran at a processor or 500 at 450Mhz and all 21164s, starting >with 13 ply out of hashtables or so. > >The 4 times faster is 100 rating points is nowadays complete nonsense. > >Speed says nothing if the conditions are met. Chess is about the weakest >point. I'm sure that bugs in diep's eval are more important than getting >another ply. > >Getting another 2 ply for free with diep definitely doesn't give me 100 >rating points. Of course within 2 years i will get up way more than >100 rating points and that's only 1 ply deeper then probably at those >hardware (considering the big heats cpu's produce nowadays i must see >that i'm 2 times faster in 2 years than what i can buy now in a shop), >but the progress 100% sure will be caused by improvements in eval and >not by that extra ply. > >>>In fact i would be amazed if there are many supercomputer programs which >>>are tactical stronger than any of the pc programs joining. > >>For many of the PC programs, it is as simple as a recompile. Supercomputers are >>IO machines. You could access the tablebase files with basically no penalty. >>That is where the biggest advantage lies, I think. > >i laugh for clusters, they can't share hashtables! > >the only accessible shared memory machines i know of are quad xeons or >8 processor Xeons, and every fast dual AMD now is not much slower as that. >it's definitely under a factor 2. > >a 16 processor 21264 or whatever you can write onto your behind, as that >machine is 10M dollar and not even Bob is getting one daily to test on the >icc! > >>>I would be even more amazed if any supercomputer based program would >>>ever win the world title (with exception from a strong PC program simply >>>running on a supercomputer; it could run on a fast pc anyway then). >> >>The CPU horsepower difference is not enormous. But if you have a huge volume of >>data, a supercomputer would be a very large advantage. > >In formula 1 racing there are teams with a budget of 50M dollar a year >and there are teams with a budget of billions of dollars a year. > >Yet both compete. > >That's how i want to compete at world champs. > >Let's play supercomputers now, before it's too late and not a single >scientist is going to get system time to join a world champ. > >Perhaps in a few years time the supercomputers are not so fast compared >to pc's, i mean using a ia64 for a supercomputer or a 400Mhz sun cpu >as basis for a supercomputer? > >Supercomputers get completely overrated when talking about computerchess! > >Further with 2 minutes a move i doubt more than 1 supercomputer will join >a world champs computerchess. > >They usually go for less rounds and more time a move, the wcc's. > >Of course they preferably would play a single round tournament! > >It's possible to win 1 game. It's impossible to win a 9 or 11 round >wmcc without being close to the best. > >>>The days that getting a ply deeper is going to win or lose a title for >>>a program are definitely over. >> >>I don't think they were ever there in the first place. With a short event with >>less than 20 matches, it's a crap shoot. But every advantage you can find >>should surely be used if you want to win. >> >>>The discussions about single or dual is quite stupid IMHO. >> >>It seems to annoy the participants, and since I am a non-participant, I >>shouldn't really have any say. But I think a discussion of fairness is always a >>good idea. > >the only valid explanation i heard from Ed which is acceptible explanation >is that rules were changed so quick before the tournament started. > >i found for this wmcc the compromise more than ok. like the gromit team >which complained at the start about there not being an amateur title i also >found this a logical complaint. > >>>Anyone, except those who still are DOS, can run his prog dual >>>using some cheapo algorithms and get a speedup that big that a cheap 1Ghz >>>dual P3 is definitely faster than a single 1.4 K7. >> >>A single SMP bug could make your program lose. > >definitely. bigger machine, more problems! >in case of supercomputer a major problem is the internet always! > >i mean considering how slow also in maastricht the internet was i definitely >wouldn't want to play at a remote machine there! > >>>For the coming years to go it's better that all those logical >>>protests from before this WMCC are going to put with the dirt outside >>>and that we concentrate upon playing a good game of chess instead >>>of complaining about those few % faster speed. >> >>If the conditions are clear, then the contestants should not fuss about it. It >>would be nice if the host university could provide strong hardware for the >>participants who cannot afford it. > >oh well the host universities in netherlands will never give you a fast >dual AMD machine, you can forget it. > >in fact they at most have a dual 300Mhz SUN or something... ...and >ICCA asks quite a bit to hire a PC... > >>>For those interested in speed, just consider that nowadays programs >>>are searching LESS deeply than the older versions of those programs >>>would do on todays hardware. >>> >>>DIEP 1997 would search like 15 plies easily at todays hardware, >>>nowadays diep searches at least 3 ply less. >> >>If the same algorithms could be pushed back to 15 plies, would you see no >>improvement? If you put your program on the 8 CPU monster system I mentioned, >>would you see no change? Don't forget that by the next WMCCC there are going to >>be Itanium machines for sure. I believe that Intel has bought some Compaq > >Itanium is a joke considering the speed it is clocked at and its >price. > >Don't tell me the joke that they might be clocked higher than a 32 >bits processor next year. 32 bits processors as long as they will exist >will be clocked higher than a 64 bits processor. > >>technology, so we might see 64 way Itanium boxes. If that comes to pass, I >>predict an everlasting butt blasting for someone who shows up on 700 MHz. > >the wcc99 was won on a single processor machine, you already forgot that? > >Please tell me about the last tournament where singles and duals and super >computers joined, where the tournament was won by a supercomputer! > >1986? > >>>wcc99 i searched 20 ply in any endgame, nowadays diep version >>>searches sometimes like 8 ply less deeply (wcc99 i ran quad xeon 400 >>>from Bob and this tournament i searched dual 1.2Ghz AMD, that processor >>>is like 25% faster as that katmai xeon processor was, not to mention >>>gcc versus visual c++ diff). >> >>Maybe you should change the name to "shallow" >>'-) > >ondiep in dutch was a name suggested by Theo v/d Storm already years ago! :) > >>>In diep's case i can easily answer why i search less deeply now, >>>especially in endgame: >>>the wcc99 version was so stupid in endgame that nearly every other >>>position gave a fail high (score high enough to not look further), >>>nowadays version though i consider it still stupid in endgame is >>>way better in endgame. less relevant is the slowdown caused by egtbs. >>> >>>I don't need to mention that the wcc99 looked like a beginner in endgame, >>>whereas the current version only against gandalf did a very stupid >>>move in endgame (Re2). >>> >>>Even a 1600 rated could easily find mistakes in 99 version of diep's >>>endgame. Nowadays such a 1600 would have a hard time finding mistakes. >> >>How will a computer do searching for them? That seems to be the question that >>matters for these contests. > >only the software matters. having software we of course try to get the >latest machine always, but also next tournament i would be amazed if someone >gets a quad. > >basically i was very happy with bob's quad and i feel in several dutch >open tournaments it mattered a lot! > >Nevertheless nowadays diep is tactically a bit more grown up, so >at 10 ply i do not fall for tricks i fell for first. > >at 12 ply i already solve a load of nolots with diep. > >i'm sure that at 12 ply i'm usual through the tactical barrier with diep. > >faster hardware is of course never bad, but it's not the weakest chain >for sure nowadays. i would be amazed if a supercomputer ever wins a >computer chess event!
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.