![[il]](/mwf/v12/flags/il.png "Israel")
from the following link in the second post
http://www.hiarcs.net/forums/viewtopic.php?t=80&start=60
"So really in our hobby/sport Intel or AMD are the truly best chess players? That blows my theory and suspicion that an engine (any engine) reputed to be the best in the world would show at its best a max 150-200 Elo improvement as a software compared to for example the Spracklen Software of 1989 inside a V10 (<---- For Steve ) if ran at the same speed"
How do I convince him that he is clearly wrong and how do I convince him that the dedicated machines with old software
simply cheat by using better hardware or by something else that is not about programming.
He gave me the following link as a proof
http://www.schachcomputer.info/html/online_aktivschach_wm_3.html
That link simply give illogical results and when I click on A I get the following results:
Res Ruffian-Elite 68060 V11 0.5-2.5
Res Deep Sjeng-Berlin Pro London 2.5-1.5
Montreux-Res Deep Sjeng 3.5-1.5
These results suggest no superiority of deep sjeng and ruffian relative to ancient software and it cannot be correct
because I know that at every time control including blitz you get clear superiority of Ruffian and commercial Sjeng
even relative to commercial programs for pc that are later commercial program then these ancient things like hiarcs7.32.
He also give me the following game when I feel that some cheating was done against rybka because my rybka2.3.2a cannot
reproduce 48...h4 regardless of depth so it seems to me that rybka did not play this game with black.
What is the truth about it and what is phoenix chess revelation rybka?
Fidelity Elite 68060 V11 (2296)
Phoenix Chess Revelation Rybka (2570)
3.Online Aktivschach WM / Abschlussturnier (6.3)
2007
1. d4 Sf6 2. c3 g6 3. Sf3 Lg7 4. e3 O-O 5. Ld3 Sc6 6. O-O d5 7. Sbd2 a5 8. b3 Dd6 9. a4 Le6 10. La3 Dd7 11. Dc2 Tfe8 12. Tae1 Lf5 13. e4 dxe4 14. Sxe4 Sxe4 15. Lxe4 Tad8 16. Lxf5 Dxf5 17. Dxf5 gxf5 18. Te2 Lf6 19. Tfe1 Kg7 20. Lc1 Kg6 21. Ld2 Td7 22. h3 h5 23. Lf4 e6 24. Kh2 Se7 25. Te5 b6 26. T5e2 Sd5 27. Ld2 Tb8 28. Kh1 c5 29. c4 Se7 30. dxc5 bxc5 31. Le3 Txb3 32. Lxc5 Sc6 33. Te3 Tb2 34. T3e2 Txe2 35. Txe2 Td1+ 36. Te1 Txe1+ 37. Sxe1 Se5 38. g3 Sxc4 39. Sd3 e5 40. Kg2 Lg5 41. La7 e4 42. Sc5 Lh6 43. Lb8 Ld2 44. Lc7 Kf6 45. Kf1 Lb4 46. Ld8+ Ke5 47. Sd7+ Ke6 48. Sf6 h4 49. gxh4 Sb2 50. h5 Lc3 51. Sg8 Sxa4 52. h6 e3 53. h7 e2+ 54. Kxe2 Lh8 55. Lf6 Sc3+ 56. Ke1 Se4 57. Lxh8 Sg5 58. Sf6 f4 59. h4 Sf3+ 1-0
http://www.hiarcs.net/forums/viewtopic.php?t=80&start=60
"So really in our hobby/sport Intel or AMD are the truly best chess players? That blows my theory and suspicion that an engine (any engine) reputed to be the best in the world would show at its best a max 150-200 Elo improvement as a software compared to for example the Spracklen Software of 1989 inside a V10 (<---- For Steve ) if ran at the same speed"
How do I convince him that he is clearly wrong and how do I convince him that the dedicated machines with old software
simply cheat by using better hardware or by something else that is not about programming.
He gave me the following link as a proof
http://www.schachcomputer.info/html/online_aktivschach_wm_3.html
That link simply give illogical results and when I click on A I get the following results:
Res Ruffian-Elite 68060 V11 0.5-2.5
Res Deep Sjeng-Berlin Pro London 2.5-1.5
Montreux-Res Deep Sjeng 3.5-1.5
These results suggest no superiority of deep sjeng and ruffian relative to ancient software and it cannot be correct
because I know that at every time control including blitz you get clear superiority of Ruffian and commercial Sjeng
even relative to commercial programs for pc that are later commercial program then these ancient things like hiarcs7.32.
He also give me the following game when I feel that some cheating was done against rybka because my rybka2.3.2a cannot
reproduce 48...h4 regardless of depth so it seems to me that rybka did not play this game with black.
What is the truth about it and what is phoenix chess revelation rybka?
Fidelity Elite 68060 V11 (2296)
Phoenix Chess Revelation Rybka (2570)
3.Online Aktivschach WM / Abschlussturnier (6.3)
2007
1. d4 Sf6 2. c3 g6 3. Sf3 Lg7 4. e3 O-O 5. Ld3 Sc6 6. O-O d5 7. Sbd2 a5 8. b3 Dd6 9. a4 Le6 10. La3 Dd7 11. Dc2 Tfe8 12. Tae1 Lf5 13. e4 dxe4 14. Sxe4 Sxe4 15. Lxe4 Tad8 16. Lxf5 Dxf5 17. Dxf5 gxf5 18. Te2 Lf6 19. Tfe1 Kg7 20. Lc1 Kg6 21. Ld2 Td7 22. h3 h5 23. Lf4 e6 24. Kh2 Se7 25. Te5 b6 26. T5e2 Sd5 27. Ld2 Tb8 28. Kh1 c5 29. c4 Se7 30. dxc5 bxc5 31. Le3 Txb3 32. Lxc5 Sc6 33. Te3 Tb2 34. T3e2 Txe2 35. Txe2 Td1+ 36. Te1 Txe1+ 37. Sxe1 Se5 38. g3 Sxc4 39. Sd3 e5 40. Kg2 Lg5 41. La7 e4 42. Sc5 Lh6 43. Lb8 Ld2 44. Lc7 Kf6 45. Kf1 Lb4 46. Ld8+ Ke5 47. Sd7+ Ke6 48. Sf6 h4 49. gxh4 Sb2 50. h5 Lc3 51. Sg8 Sxa4 52. h6 e3 53. h7 e2+ 54. Kxe2 Lh8 55. Lf6 Sc3+ 56. Ke1 Se4 57. Lxh8 Sg5 58. Sf6 f4 59. h4 Sf3+ 1-0
![[nl]](/mwf/v12/flags/nl.png "Netherlands")
Hi Uri,
I am always interested to see how old programs behave on modern hardware to "measure" the improvement of chess engine since the 80's or the 90's.
But here you address the "reverse" question: How would behave the new engine on old hardware compared to old program designed to run on those old hardware. This is a less interesting question from a pratical point of view but it is interesting anyway.
I think it is possible that indeed the gap in the rating would be smaller. Indeed takes the extreme case where the hardware is so slow that you can only compute up to depth 1 (without additional sellective search), so the program search only for his next move. In this set up, chess becomes purely random and luck is the main ingredient. In this case if you take two engines and let them play a LARGE number of game, the result should be equal.
Now in the less extreme case where you can reach depth 4 or 5 then it is not a surprise for me if the gap between to engine would decrease because there is still a part of luck with such low depths. Moreover the new engine are designed to try to go a deep as they can in the tree, perhaps they use less knowledge than before for that, so they would be less strong on slow hardware where they cannot reach a large depth where they start to become really strong, I do not know...
I do not know if one can do it under Arena but if I can I will try to make a match between several engines by limiting the depth to 4 or 5. The games would be very fast on modern hardware and this would be a test to see if the gap in the rating decreases.
Regards,
Fabian
I am always interested to see how old programs behave on modern hardware to "measure" the improvement of chess engine since the 80's or the 90's.
But here you address the "reverse" question: How would behave the new engine on old hardware compared to old program designed to run on those old hardware. This is a less interesting question from a pratical point of view but it is interesting anyway.
I think it is possible that indeed the gap in the rating would be smaller. Indeed takes the extreme case where the hardware is so slow that you can only compute up to depth 1 (without additional sellective search), so the program search only for his next move. In this set up, chess becomes purely random and luck is the main ingredient. In this case if you take two engines and let them play a LARGE number of game, the result should be equal.
Now in the less extreme case where you can reach depth 4 or 5 then it is not a surprise for me if the gap between to engine would decrease because there is still a part of luck with such low depths. Moreover the new engine are designed to try to go a deep as they can in the tree, perhaps they use less knowledge than before for that, so they would be less strong on slow hardware where they cannot reach a large depth where they start to become really strong, I do not know...
I do not know if one can do it under Arena but if I can I will try to make a match between several engines by limiting the depth to 4 or 5. The games would be very fast on modern hardware and this would be a test to see if the gap in the rating decreases.
Regards,
Fabian
Hi Fabien,
Of course in this extreme case the difference is going to be very small but still the claim of difference of 200 elo does not make sense even if we use an old hardware.
The speed that he talked about is 200 mhz and not 0.01 mhz or 1 mhz and I see no way that with 200 mhz you get only depth of 4 plies.
rybka can easily get 10 plies even with 200 mhz at 30 seconds per move that is the time control that he talks about because on my hardware it often get 10 plies or more than it in less than a second.
Uri
Of course in this extreme case the difference is going to be very small but still the claim of difference of 200 elo does not make sense even if we use an old hardware.
The speed that he talked about is 200 mhz and not 0.01 mhz or 1 mhz and I see no way that with 200 mhz you get only depth of 4 plies.
rybka can easily get 10 plies even with 200 mhz at 30 seconds per move that is the time control that he talks about because on my hardware it often get 10 plies or more than it in less than a second.
Uri
Yes it seems also to me that at 200 mhz you should reach depth high enough such as the new engines (Rybka in particular) would stay well above the old ones.
I do not have anymore such an old hardware to test, it would be interesting but not so useful from a practical point of view. We have new hardware we (not me :) ) have design engine that play well on those hardware and not on old hardware...
I do not have anymore such an old hardware to test, it would be interesting but not so useful from a practical point of view. We have new hardware we (not me :) ) have design engine that play well on those hardware and not on old hardware...
![[fi]](/mwf/v12/flags/fi.png "Finland")
> I do not know if one can do it under Arena but if I can I will try to make a match between several engines by limiting the depth to 4 or 5. The games would be very fast on modern hardware and this would be a test to see if the gap in the rating decreases.
The practical meaning of depth varies wildly between different engines. Try extremely short time limits instead; you can use 0 seconds + an increment of a few milliseconds (the book moves will ensure the engines don't lose on time). Some (most?) engines are going to have trouble, but it's easy to take any open source engine and modify it to check the remaining time often enough not to lose on time.
Node limits are also a possibility if you use engines with similar nps.
BTW, someone posted the results of a 1-ply tournament here a few months ago.
Engine Score Ry Ry Ry Ry Ry Kt Gl Hi Ph Fr Na Sp To De Sc De Ru SO Cr Lo Sm S-B
01: Rybka 2.2n2 37.0/40 ··· 10 1= 11 11 10 11 11 =1 11 11 11 11 11 11 11 11 11 11 11 11 670.00
02: Rybka 2.1c 37.0/40 01 ··· 11 11 01 01 11 11 11 11 11 11 11 11 11 11 11 11 11 11 11 666.50
03: Rybka 2.3lk 34.0/40 0= 00 ··· =1 11 11 10 10 11 11 11 11 11 11 11 11 11 11 11 11 11 579.00
04: Rybka 2.3 33.0/40 00 00 =0 ··· 10 11 11 11 11 11 11 1= 11 11 11 11? 11 11 11 11 11 535.25
05: Rybka 1.0 Beta 31.5/40 00 10 00 01 ··· 11 =0 11 11 11 == 11 11 11 11 11 11 11 11 11 11 510.75
06: Ktulu 8.0 31.0/40 01 10 00 00 00 ··· 10 11 11 11 11 11 11 11 11 11 11 11 11 11 11 482.50
07: Glaurung 24.5/40 00 00 01 00 =1 01 ··· =1 10 =1 11 10 11 11 01 01 0= 11 1= 11 11 385.25
08: Hiarcs 11 22.5/40 00 00 01 00 00 00 =0 ··· == 11 10 11 11 =0 11 1= 11 01 11 11 11 306.75
09: Pharaon 22.0/40 =0 00 00 00 00 00 01 == ··· 10 =1 1= =1 11 11 00 11 11 11 11 11 290.00
10: Fruit 2.2.1 18.5/40 00 00 00 00 00 00 =0 00 01 ··· 00 11 =1 =1 10 11 11 11 10 11 11 217.50
11: Naum 17.5/40 00 00 00 00 == 00 00 01 =0 11 ··· 00 01 10 1= =0 11 11 11 == 11 228.00
12: Spike 1.2 16.5/40 00 00 00 0= 00 00 01 00 0= 00 11 ··· 01 10 0= 11 == 1= =1 11 11 201.00
13: TogaII 16.0/40 00 00 00 00 00 00 00 00 =0 =0 10 10 ··· =1 10 =1 1= 11 1= 11 11 171.75
14: Deep Shredder 10 15.5/40 00 00 00 00 00 00 00 =1 00 =0 01 01 =0 ··· 01 11 10 11 11 10 11 179.50
15: Scorpio 15.5/40 00 00 00 00 00 00 10 00 00 01 0= 1= 01 10 ··· 10 11 01 =1 11 11 177.25
16: Delfi 13.5/40 00 00 00 00? 00 00 10 0= 11 00 =1 00 =0 00 01 ··· 00 10 11 11 11 164.50
17: Ruffian 12.5/40 00 00 00 00 00 00 1= 00 00 00 00 == 0= 01 00 11 ··· 11 0= 11 11 135.00
18: SOS 5.1 9.0/40 00 00 00 00 00 00 00 10 00 00 00 0= 00 00 10 01 00 ··· 11 1= 11 83.50
19: Crafty 21.5 8.5/40 00 00 00 00 00 00 0= 00 00 01 00 =0 0= 00 =0 00 1= 00 ··· 11 11 82.50
20: LoopList 4.5/40 00 00 00 00 00 00 00 00 00 00 == 00 00 01 00 00 00 0= 00 ··· 11 37.50
21: SmarThink 0.0/40 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ··· 0.00
Yes but as you point out, one has to be sure that 1-ply means the same for each engines... In such a tournament if one engine goes always up to ply 2 it is enough to win many more games... So certainly it is difficult to "simulate" old hardware with our new ones...
I am running a tournament 0 sec + 0.5 sec/move (10 moves from the common opening book, so they start with 5 sec + increment) beetween Rybka 2.3.2a, Hiarcs 11.2, Loop 13.6 and Shredder 10.1. I will post here the results and the corresponding ratings and compared them to ratings obtained from tournaments with 1min + 1sec/move and tournament with 40 moves/5 minutes I am running regularly on my website (http://www.ordichec.net).
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill