Author: Uri Blass
Date: 19:42:53 01/03/01
Go up one level in this thread
On January 03, 2001 at 18:02:43, Bertil Eklund wrote: >On January 03, 2001 at 17:06:58, Thorsten Czub wrote: > >>On January 03, 2001 at 10:18:51, Bertil Eklund wrote: >>>>Then it's even worse for program B. In this case, "B benefits more from faster >>>>hardware" is a statement created to hide the fact that B sucks somewhere on the >>>>strength/time_control curve. Generally near the origin (fast time controls). >>>> >>>>The "benefits more from faster hardware" is a bullshit. It's an attempt to make >>>>you believe that the program in question will be the best on the hardware of >>>>next year. Which never happens. >> >> >>nonsense. there is a normal reason the programs on fast hardware do >>not beat next year, because in 1 year the others make also much progress. >>they buy the program that would be strong on fast hardware, and >>next year the slow program has not much chance. >> >>saying chess programs would increase strength linear, is a myth ! >>it's not true. >>not for fritz (strong in first seconds, weaker the deeper) >>not for hiarcs (the more time the better) , not for shredder, >>nor mchess nor cstal (the faster or more time control the better) >>and and and and. >> >>these programs all do not increase strength in linear graphs, >>your problem is: that you don't feel or measure or find out >>why and when chess programs play strong. >>with ssdf you only measure a fracture of what you could measure >>if you would study main-lines and search-depth and would watch the >>games all and all live. >> >>take e.g. gandalf. >>this program develops into a monster on 1000 Mhz hardware on 40/120. >>if you play faster than 3' per move , or on a slower machine, >>others are stronger. >> >>but i cannot guaranty you, that in the time ssdf-tests with 1000 Mhz >>machines, gandalf432 will be the leader, because 1 or 2 years >>later, this is 10 or 20 years in computerchess. >>the relation between developments between cars and computers >>is nearly 1:5. >>4 years in computer-time is nearly 20 years in car-devlopment. >> >>gandalf is as good as example as hiarcs or cstal is. >>but when the time comes, you test all programs on THAT hardware, >>the others have made big progress too. >> >>the progress in gandalf/hiarcs/cstal came from intelligence in those >>programs. >>it takes time until other programs get this intelligence too. >>but when the time has come, there is no advantage anymore. >> >>>>It has been said many times for various programs, but then it has never been >>>>proven. >> >>>> Christophe >> >>but this sentence is not true. >>you cannot prove it. because when YOU find out, or the ssdf, thats so late >>that you have forgotten how strong something was months ago. >> >> >>>Hi! >>> >>>In fact it is one case CStal competes relatively good on "todays" computers. >>> >>> >>>Bertil >> >>hiarcs, gandalf, old mchess versions. > >I see that your answer was mainly to Christophe but in the case of the latest >Hiarcs7.32 it's probably weaker on longer time-controls. Hiarcs is an excellent >blitzer but slightly weaker on longer time-controls. > >This is probably because of the search it uses, in example when Hiarcs goes one >ply deeper it often takes 10 times longer than the last ply and therefore cant >finish the next ply. > >Bertil Not only because of this but also because of the retaining hash tables bug that is a practical problem only at long time control. 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.