Author: Robert Hyatt
Date: 17:43:01 01/03/00
Go up one level in this thread
On January 03, 2000 at 18:21:53, Bertil Eklund wrote: >On January 03, 2000 at 16:53:35, Robert Hyatt wrote: > >>On January 03, 2000 at 03:25:35, Bertil Eklund wrote: >> >>>On January 02, 2000 at 07:08:32, Rajen Gupta wrote: >>> >>>> >>>> >>>> >>>>>>> Doubling the speed of the engine is supposed to produce an increase of about >>>>>>>60 elo points. >>>>>> >>>>>>I do not think that this assumption is right against humans. >>>>>> >>>>>>Uri >>>> >>>>>> >>>>>Why? >>>>> >>>>>Bertil >>>>Hi all;i think the answer is best summed up by what bob hyatt has been saying >>>>all along-ie a chain is only as strong as its weakest link and in the case of >>>>computer chess this weak link consists of lack of long range and startegic >>>>planning and lack of positional understanding-once this weak link is identified, >>>>it can be attacked by a good enough chess player anfd further increases in >>>>computer hardware wouldn't make much difference.in computer vs computers on the >>>>other hand are playing to each others strenghths rather than weaknesse so >>>>hardware increse would show up significantly-i think this has been pretty clear >>>>from the inability of rebel to have a plus score against grandmasters >>>> >>>>rajen >>> >>>Hi! >>> >>>Mr Hyatt has already answered on this with two tongues, yes. He admits though, >>>that his Crafty on quad Xeon is better than his Scrappy on quad Pentium Pro. >>> >>>When did it happened that the increase of speed stopped programs from playing >>>better? With the step from 286 to 386, from 386-486 or is it the step from >>>Pentium to Pentium2 or perhaps Deep Thought on one, four or sixteen cpu´s. >>> >>>Bertil >> >> >>I don't answer "with two tongues". I have said _many_ times, "computer vs >>computer games tend to exaggerate the rating difference between the two programs >>when only one thing is changed." It has been common knowledge that a machine >>twice as fast will enjoy a <roughly> 60-70 point advantage over the _same_ >>program on the original (1/2 speed) hardware. This number was not a guess. It >>was gleaned from years of testing on faster and faster hardware by many >>different people... From the SSDF to IM Larry Kaufman. But it is only valid >>in the context it was tested, machine vs machine. There is _nothing_ that >>suggests that 2x hardware is 60-70 Elo points better against _human_ players. >> > >A doubling of the speed gave about 60-70 Elo in the past with 486 and Pentium >this was more or less proved in those days, and I guess it is still the same in >blitz/speed-games against humans. When was the day (cpu) when this formula >stopped working against humans? Was it the same day as Aegon began with >increments or humans play one game matches against computers (with >double-increments). It didn't "stop working". It "never worked". The data has never been produced. No one has taken a stable group of human players and then tried (say) program X on a Pentium 60, then a 120, then a 240, then a 480, to see the difference. Everyone has been much more concerned/excited about program vs program comparisons. > >BTW a P800 vs a P400 is only 60-70% faster. depends. For Crafty, a P6/200 vs a xeon/400 is pretty much 2.0 faster. > >>Such data is more difficult to obtain because no one plays that many games and >>records the results. But I agree with Uri. Against a GM, a program that would >>have a fide rating of X on a PII/400 would not have a rating of X+60 on a >>PII/800 processor. In many cases the improvement will be zero, because often >>depth is not the only issue. Knowledge comes into play, and depth doesn't >>always equate with more knowledge. > > Yes over a single game the improvement can be zero, if the programs evaluation >is totally wrong it can be weaker, but if the zero-case is true why don´t play >the same game on half the time. > >>We may never know what doubling does vs GM players because we need a lot of data >>and it is not coming very quickly. > >Absolutly true. > >Bertil
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.