Author: Chris Carson
Date: 05:41:11 07/25/00
Go up one level in this thread
On July 25, 2000 at 01:40:29, Alvaro Polo wrote: >On July 25, 2000 at 01:34:41, Alvaro Polo wrote: > >>On July 25, 2000 at 00:07:33, Christophe Theron wrote: >> >>>On July 24, 2000 at 16:30:39, blass uri wrote: >>> >>>>On July 24, 2000 at 15:59:31, Chris Carson wrote: >>>> >>>>>The SSDF Rating List 1997-02-05 >>>>>53540 games played by 162 computers >>>>> Rating +- Games Won Oppo >>>>> ------ --- ----- --- ---- >>>>> 1 Rebel 8.0 Pentium 90 MHz 2462 27 736 67% 2336 >>>>> 2 MChess Pro 6.0 Pentium 90 MHz 2435 27 683 62% 2350 >>>>> 3 Hiarcs 5.0 Pentium 90 MHz 2427 67 121 60% 2359 >>>>> 4 Genius 5.0 Pentium 90 MHz 2420 30 558 59% 2352 >>>>> >>>>> >>>>>Avg Rating for the top 4 programs = 2436. >>>>> >>>>>If DB had a 90% (+36 =4 -0) score against these programs, >>>>>then the DB rating in comp vs comp games would have been 2824. >>>>>It's TPR against Kasparov was 2862 (human vs comp). >>>> >>>>+36 =4 -0 score is 95% score and not 90% score. >>>> >>>>I did not read a claim that Deeper blue did +36 =4 -0 against these programs. >>>> >>>>I remember a claim that Deep blue Junior That is weaker than Deep blue did 38:2 >>>>score against programs but I do not know exactly the name of the programs and it >>>>is better for Hsu not to tell the names of the programs because it is better to >>>>say nothing when you have no proof. >>>> >>>>I know that Deep blue Junior lost 3:0 against Rebel and drew 1.5-1.5 against >>>>Rebel-tiger. >>> >>> >>> >>> >>>You should not give only part of the information. If you quote a match result, >>>you should give as much information as possible. >>> >>>The games took place in the hall near the tournament hall were the 1999 WCCC >>>took place, in Paderborn, in June 1999. >>> >>>There were several PCs connected to Deep Blue Junior thru an Internet >>>connection. Deep Blue Junior was in free access and I'm very surprised that I >>>have been the only one to try to play with my program against the monster. >>> >>>I could not bring the 600MHz Kryotech computer I was using for the regular >>>tournament. It was only 20 to 30 meters away from the tournament hall, but the >>>Kryotech computer is rather heavy, and anyway there was no power supply >>>(actually that's what I thought - later I discovered a power supply nearby and >>>used it). >>> >>>So I brought my Pentium 150MHz laptop and decided to try some games. I thought I >>>was going to be totally destroyed anyway, so it was just for fun. >>> >>>Deep Blue Junior was always using 1 second per move. >>> >>>I set up Chess Tiger 11.9 (Paderborn version) to play the game in 15 minutes. >>>Permanent brain was turned OFF. The reason to use this setup is that I wanted to >>>see what could happen if Chess Tiger was using a top level hardware of year >>>2000. I thought that such hardware would probably be 15 times faster than my >>>P150 notebook. And that DBJr was not using permanent brain. >>> >>>Why did I do that? At that time I had heard rumours that Hsu was going to >>>release a PC card with one of the DB chips on it. I wanted to know if it was >>>possibly going to make the top PC programs look ridiculous. >>> >>>Chess Tiger was using 8Mb hash tables. >>> >>>Here are the two games played on the P150 Notebook: >>> >>> >>>[Event "during the WCCC99"] >>>[Site "Paderborn, Germany, via Internet"] >>>[Date "1999.06.17"] >>>[Round "?"] >>>[White "Chess Tiger 11.9"] >>>[Black "Deep Blue Junior"] >>>[Result "*"] >>> >>>1. e4 d5 2. exd5 Nf6 3. d4 Nxd5 4. c4 Nf6 5. Nf3 e6 6. Nc3 c5 7. d5 exd5 >>>8. Nxd5 Be7 9. Nxf6+ Bxf6 10. Qxd8+ Bxd8 11. Be3 Be7 12. Bd3 Nc6 13. O-O Nb4 >>>14. Be4 f5 15. Bb1 Bf6 16. Re1 Kf7 17. Bc1 Re8 18. Rxe8 Kxe8 19. a3 Nc6 >>>20. Bc2 Na5 21. Be3 Bxb2 22. Re1 Kf7 23. Bxc5 Nxc4 24. Bd3 Nxa3 25. Re2 Bc1 >>>26. Nd4 b6 27. Bd6 a5 28. Ra2 Be6 29. Nxe6 Kxe6 30. Bxa3 Bxa3 31. Rxa3 Rc8 >>>32. Kf1 g6 33. Ke2 Rc6 34. Kd2 Rd6 35. Rb3 h6 36. Kc3 * >>>(the game was stopped here due to an operator mistake. >>>DB Jr does not allow any takeback. But the position looks like a win >>>for Chess Tiger, whose score was at that time over +2.00) >>>(time control was 1s/move for DB Jr, and 15mn/game for Chess Tiger. >>>Chess Tiger ran on a P150 notebook in a W95 Dos box, with 8Mb hash tables) >>> >>>In this game Tiger was out of book after move 5 (Nf3). >>> >>> >>> >>> >>>[Event "during the WCCC99"] >>>[Site "Paderborn, Germany, via Internet"] >>>[Date "1999.06.17"] >>>[Round "?"] >>>[White "Chess Tiger 11.9"] >>>[Black "Deep Blue Junior"] >>>[Result "0-1"] >>> >>>1. e4 e5 2. Nf3 Nc6 3. Nc3 Bb4 4. Nd5 Nf6 5. Nxb4 Nxb4 6. c3 Nc6 7. d4 exd4 >>>8. e5 Qe7 9. cxd4 d6 10. Bb5 dxe5 11. Bxc6+ bxc6 12. dxe5 Ba6 13. Qa4 Bb5 >>>14. Qh4 Qe6 15. h3 O-O 16. Be3 Nd5 17. Bc5 Rfb8 18. Qe4 Nf6 19. Qe3 Nd5 >>>20. Qe4 Nf6 21. Qc2 Rd8 22. Be3 Bd3 23. Qc3 Ne4 24. Nd4 Nxc3 25. Nxe6 fxe6 >>>26. bxc3 Rab8 27. Bc1 Bc4 28. a3 Rb3 29. h4 Rxc3 30. Rh3 Rc2 31. Re3 Rd4 >>>32. g3 c5 33. a4 Bd5 34. a5 c4 35. Ba3 Rdd2 36. Rb1 h5 37. Bc5 Rxf2 >>>38. Rc3 Rfe2+ 39. Kd1 Rcd2+ 40. Kc1 Ra2 41. Rb8+ Kh7 42. Rb2 Rexb2 >>>43. Ra3 Rc2+ 44. Kd1 Rd2+ 45. Ke1 Re2+ 46. Kd1 Rad2+ 47. Kc1 Rc2+ >>>48. Kd1 Red2+ 49. Ke1 Rh2 50. Kd1 Rcd2+ 51. Kc1 Rdg2 52. Bf2 Rxf2 >>>53. Rd3 cxd3 54. Kd1 Rf1 0-1 >>>(time control was 1s/move for DB Jr, and 15mn/game for Chess Tiger. >>>Chess Tiger ran on a P150 notebook in a W95 Dos box, with 8Mb hash tables) >>> >>>In this game Tiger was out of book after move 3 (Nc3). >>> >>>OK, this time Tiger gets its spanking. >>> >>> >>> >>>The third game was played the next day, after Ed came with his PII-333MHz >>>Notebook (he is obviously richer than me :). I borrowed the PII-333 Notebook and >>>played the third game: >>> >>> >>>[Event "during the WCCC99"] >>>[Site "Paderborn, Germany, via Internet"] >>>[Date "1999.06.18"] >>>[Round "?"] >>>[White "Chess Tiger 11.9"] >>>[Black "Deep Blue Junior"] >>>[Result "*"] >>> >>>1. e4 d5 2. exd5 Nf6 3. d4 Nxd5 4. c4 Nf6 5. Nf3 e6 6. Nc3 c5 7. d5 exd5 >>>8. Nxd5 Nxd5 9. cxd5 Be7 10. Bb5+ Nd7 11. O-O O-O 12. Bf4 Qb6 13. Qd3 a6 >>>14. Bc4 Qxb2 15. Qe4 Bf6 16. Rab1 Qa3 17. d6 Rb8 18. Bc1 Qa5 19. Bd3 g6 >>>20. Bh6 Bg7 21. Bxg7 Kxg7 22. Qe7 Qxa2 23. Ne5 Qd5 24. Nxd7 Bxd7 25. Be4 Qe6 >>>26. Qxe6 fxe6 27. Rxb7 Rxb7 28. Bxb7 Bb5 29. Ra1 Rb8 30. d7 Bxd7 >>>31. Bxa6 Ra8 32. Rd1 Ba4 33. Rd6 Bb3 34. Rb6 Bd5 35. f3 c4 36. Kf2 c3 >>>37. Ke3 Ra7 38. Kd3 Rc7 39. Kc2 Kf6 40. Rb4 g5 41. Bd3 h6 42. h4 Rc5 >>>43. Rg4 Ke7 44. hxg5 hxg5 45. Ra4 Kf6 46. Ra3 Ke5 47. Ra4 Rc7 48. Rg4 Kf6 >>>49. Ra4 Rc5 50. Ra3 Ke5 51. Ra4 Rc8 52. Rg4 Kf6 53. Ra4 * >>>(here the game was stopped because the notebook's battery was exhausted >>>and the notebook turned itself off automatically. It has not been possible >>>to resume the game because we needed some time to charge the battery and the >>>tournament hall was about to close. >>>It is not clear what the outcome of the game could be, but it looked like >>>DB Jr was unable to improve its position) >>>(time control was 1s/move for DB Jr, and 10mn/game for Chess Tiger. >>>Chess Tiger ran on a PII-333 notebook in a W95 Dos box, with 8Mb hash tables) >>> >>> >>>So the result of the match is 1.5-1.5 after adjudication, but I agree it could >>>have been 2-1 in favor of DB Jr. >>> >>>I must add that there have been at least 2 (or was it 3?) other games played >>>with the P150 Notebook, but I did not save the games because each time I did an >>>operator mistake very early (just after the end of the opening). >>> >>>You have to understand that DB Jr did not allow any takeback. So in case of a >>>mouse slip, which happens too often, it was impossible to continue. I was very >>>anxious during this small match, and that is the reason behind all these >>>mistakes. >>> >>>I was anxious because I did not expect to win ANY game. But as you see, Tiger >>>simply won the first game... >>> >>>So now let's see what happened. On his hardware, Tiger was computing only 25,000 >>>positions per second. At 15mn/game time control, that means it was computing >>>375,000 positions per search in average. >>> >>>Isn't Deep Blue supposed to compute way faster? I don't remember the numbers. >>>Was it 1M nodes per second per chip, or 2M nodes per second? >>> >>>If it's only 1M nodes per second and it could only use 3/4 of a second for its >>>search (the rest being taken by "downloading stuffs into the chip" as Bob said), >>>then it's still 750,000 positions per search, twice the number of positions that >>>Tiger could compute during its search on P150. >>> >>> >>>So my conclusion is that I have seen nothing special in this match. I have seen >>>2 chess programs fighting, the one computing more nodes taking the advantage, >>>but certainly not crushing its opponent as some people would like us to believe. >>> >>>The funny thing is that before playing the match I thought I would be crushed. >>>You see, I have been the victim of the propaganda myself... >>> >>> >>>Now if you ask me about the chances of Chess Tiger against Deeper Blue and its >>>200 processors at tournament time controls, I simply say that I think that Chess >>>Tiger has absolutely no chance. >>> >>>But against a single chip, I would say that a program like Chess Tiger running >>>on current top hardware has its chances. >>> >>>Remember that in similar circumstances (fast games played in the same hall) >>>Rebel won against Deep Blue Junior by 3-0. >>> >>>And you know what? Given that Deep Blue does no forward pruning, this is NO >>>SURPRISE. >>> >>> >> >>This is really interesting. Tiger at 375,000 nps against DBJ at 750,000 nps and >>obtaining 1.5-1-5 or possibly 1-2 suggests that DBJ was really DBJ and not some >>crippled version running under 2200 Elo. It also suggests that the full 480 >>processor DB, at 200M nps, if CT has a comp-comp rating of nearly 2650, would be >>over 3200 Elo comp-comp (8 doublings of speed at 70 Elo each doubling) >> >>Alvaro >> > >Sorry, this calculation is wrong. I misread and confused 375,000 positions per >search with nodes per second. If would be 560 Elo above the 375,000 pos per >search CT, not the SSDF tested Chess Tiger. > >Alvaro Not a bad estimate, however, it should be noted that 560 over a 150mhz pentium is not that much (aprox. 7 times at 80 pts/double). A p150 has a specint2000 rating of about 40, so a specint2000 of 2560 would be required. The p3-700 has a specint2000 of 310, so 8x-700 is 1240 or within a factor of 2 (one doubling). Considering the human vs computer results, this is not a bad estimate. However, diminishing returns would be in effect for both, crafty gets a speed up of 3/4 for 4 processors and I think DJ got a speed up of 4-5 for the 8 processors. So both the 560 points and the needed specint2000 may be inflated. This means that the single processor programs/machines may be closer than they appear (using analysis) for single vs multi-processor hw. This just gives me more confidence that the programmers for Rebel, Chess Tiger, Deep Junior, and Fritz have made and continue to make great improvements in there eval and search algorithms to get so much more out of the hardware they target. :) Best Regards, Chris Carson >>> >>> Christophe
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.