Author: Ricardo Gibert
Date: 17:55:04 01/31/02
Go up one level in this thread
On January 31, 2002 at 11:23:35, Ricardo Gibert wrote: >On January 31, 2002 at 08:38:31, Uri Blass wrote: > >>On January 31, 2002 at 08:33:32, Ricardo Gibert wrote: >> >>>On January 30, 2002 at 12:11:16, Robert Hyatt wrote: >>> >>>>On January 30, 2002 at 11:00:58, Alexander Kure wrote: >>>> >>>>>On January 30, 2002 at 10:25:41, Robert Hyatt wrote: >>>>> >>>>>[..] >>>>> >>>>>>You can make up all the math you want, but it doesn't prove anything. I >>>>>>_know_ that DB's branching factor was roughly 4.0, as was discussed _here_ >>>>>>a few years ago after several of us looked carefully at their logs. >>>>>> >>>>>>to go to depth 18 requires 4^17 as many nodes as searching to one ply. >>>>>>4^17 = 2^18, = 262,000 roughly. >>>>> >>>>>4^17 = 2^34 >>>>> >>>>>[..] >>>>> >>>>> >>>>>Greetings >>>>>Alex >>>> >>>> >>>>You are right. Wasn't thinking clearly at the time, obviously. >>>>:) >>>> >>>>Bob >>> >>>But at the time, you must have thought you were thinking clearly or you surely >>>would not have made the post. This raises the question of, "How do you know you >>>are thinking clearly now?" ;-) >>> >>>Nothing "obvious" about it, yes? >>> >>>Setting aside my stupid jokes, the serious question now is: "Isn't 2^34 a bit >>>too big for Deep Blue?" >> >>200M nodes/second*180 seconds=36*10^9>2^34 > >2^34 figure is how many times more nodes must be searched than a 1 ply search >(see http://www.icdchess.com/forums/1/message.shtml?210884). If we make the very >generous assumption that the BF at the root is only 4 (at least 30 actually. RH >assumed 100 for example), then the inequality becomes 3.6E10 < 6.4E10. Actually, after reading what RH wrote in rgcc, 100 for the root seems to be the appropriate number to use rather than 4. Ouch! So multiply by a factor of 25. If we assume peak NPS of 1G, then only multiply by 5. Curiously, RH casts off the discrepancy by suggesting that the last iteration is incomplete, but this can at most account for a factor of 4. And remember, we are still discounting what iterative deepening adds to the count in number of nodes searched. > >When you factor in that this assumes iterative deepening is not employed, then >it is not even close. I will admit that it is closer than I originally thought, >however. > >> >>2^34 nodes is not too big for a machine that can calculate 200M nodes per >>second. >> >>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.