Author: Robert Hyatt
Date: 07:37:35 09/18/05
Go up one level in this thread
On September 18, 2005 at 01:07:07, Ryan B. wrote: >On September 18, 2005 at 00:34:50, Robert Hyatt wrote: > >>On September 18, 2005 at 00:28:19, Ryan B. wrote: >> >>>On September 17, 2005 at 23:49:59, Robert Hyatt wrote: >>> >>>>On September 17, 2005 at 23:32:46, Ryan B. wrote: >>>> >>>>>On September 17, 2005 at 21:53:26, Robert Hyatt wrote: >>>>> >>>>>>On September 17, 2005 at 12:49:06, Ryan B. wrote: >>>>>> >>>>>>>I thought this rediculess theory was proven wrong with the Crafty vs Rebel thing >>>>>>>a long time ago. It is far more complicated than just the NPS loss in move >>>>>>>ording, pruning, reduction checks, and extention checks that all can improve a >>>>>>>program while reducing the NPS. As well what is known by the eval function and >>>>>>>how it is used can be much more valuable than even another few ply searched but >>>>>>>causing a very large loss in NPS. Hiarcs on slow hardware vs GNU Chess on fast >>>>>>>hardware should show a good example of this. >>>>>> >>>>>> >>>>>>What was proven wrong? >>>>>> >>>>>>We played exactly one game with crafty vs rebel, with the time handicap. Ed >>>>>>later played another match with a _completely_ different result (his chess 2010 >>>>>>or whatever it was. >>>>>> >>>>>>I'd say that shredder is clearly better than current crafty. Anyone want to >>>>>>give me 1000:1 time odds and play a match? >>>>>> >>>>>>Just name the when/where... >>>>>> >>>>>>I _know_ how such a match will turn out... >>>>> >>>>>Take null move out and I know the result as well. You might get shocked.... >>>>>again. >>>> >>>>I've not been shocked the _first_ time yet. >>>> >>>>And I also have no idea what null-move has to do with this. Should I also take >>>>out search extensions? The evaluation? Endgame tables? Should I just play >>>>using a text editor? >>>> >>>>Again, two programs that are reasonable, if one goes way faster, it will win way >>>>more. And nothing is going to shock that out of me... >>> >>> >>>How could the Rebel vs craft game even if just one game not shock you? Also I >>>would like to know why a person with your experiance and enducation would say >>>"if one that goes way faster" knowing "way faster" can be defined in many >>>differant ways and you are destine to be taken out of context with such a >>>comment. >> >>10x faster is a significant advantage. 100x is more significant. 1000x is more >>significant. What is ambiguous about that? >> >> >> >>> If you mean much faster hardware than you have a solid but not >>>flawless statment >> >>This has _always_ been about faster hardware. Not lobotomizing a chess engine >>which would make absolutely no sense in any sort of "scientific test". >> > >Good to know we are on the same page here. "IF" NPS was all that matterd >lobotomizing a chess engine would help its play. I think we should all know >this is not the case. This often gets twisted into that, because Vincent always chimes in with "DB's eval was like gnuchess". With no proof. No supporting evidence. Even though their presentations on the "auto-tuner" mentions tuning 8000 parameters. GNU didn't have 800 paramenters, much less 8000. So the thread always get side-tracked into something that can never be tested. Which is safe ground for those wanting to make outlandish statement of course... The experiment I proposed can easily be done... From where I sat a month ago, Crafty looked pretty competitive running at 16M nodes per second. How competitive would it be running at 16B nodes per second? No program will win every game, even at that speed, but it sure would not lose many, and it would certainly win a bunch... > >> >>> If you mean much higher NPS your statment is simply wrong. >> >> >>Nope. Faster hardware _always_ equals faster NPS for a given program. So it >>isn't "simply wrong". If you want to somehow imagine that I said that taking >>two programs on the same computer, and the faster NPS program will win, then you >>are free to do so. But I certainly never said, nor implied such a thing. Feel >>free to dig up a context-complete quote where I did, of course, although I know >>such a thing can't be found. >> > >I did not know we where talking about the same program vs itself on differant >hardware. We were not necessarily talking about that. The idea was "reasonably similar programs". You could choose two programs, one that is significantly stronger than the other, so long as they were reasonably "similar". 1000x will completely erase that "significantly stronger" and turn that around dramatically... > >> >>>If you mean something else please explain. >> >> >>I meant exactly what I said. > >And it was worth clerifying, I felt what I read was unclear. People will take >you out of context if you let them. Here, all the time. :)
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.