Author: Robert Hyatt
Date: 09:29:13 08/30/01
Go up one level in this thread
On August 30, 2001 at 12:03:59, Mark Young wrote: >On August 30, 2001 at 11:33:58, Robert Hyatt wrote: > >>On August 30, 2001 at 10:51:23, Mark Young wrote: >> >>>On August 30, 2001 at 10:29:35, Mogens Larsen wrote: >>> >>>>On August 30, 2001 at 10:20:12, Mark Young wrote: >>>> >>>>>You will answer it for my Bob! Here is your Quote and the full text. >>>> >>>>Nothing in the quoted text supports your own "little progress" interpretation. >>>>There is "small number of "revolutionary" ideas", "slow, methodical progress" >>>>and "incremental changes". All of which underlines slow and steady quite well >>>>according to my understanding of the English language. But what do I know, I'm a >>>>foreigner. >>> >>>I will Quote again for the foreigners: >>> >>>"Part of the progress has been due >>>to incremental changes to chess engines/evaluations/etc, part has been due to >>>the hardware speed advances. Probably more of the latter than the former, if >>>the truth is known..." >>> >>>Bob states clearly from the above that he thinks "if the truth is known" Micro >>>Chess computer advancement is more due to hardware speed or faster computers. >>>Less to due with better chess programs. >> >> >>You keep quoting the same text, and you keep avoiding the question that I >>(and now others) have asked. Here it is again: >> >>-------------------------------------------------- >>where did I say "little software progress has been >>made over the past 10 years."??? >>-------------------------------------------------- > >Im glad this is not what you mean in your statements...as software is much >stronger today then even a few years ago...disregarding hardware. I still >disagree with slow and steady, as I think we can show fast progress on the >software side. I am not sure where the "fast" progress has been. IE I don't see any totally new search algorithm (including parallel search which has been around for well over 20 years already), any new anything really. The programs of today fit in the mold of chess 4.x, with a few enhancements thrown in here and there. Even the forward-pruning stuff was around in the days of Greenblatt. There are things we can do today that we could not do 10 years ago, because back then they would have been too costly and would have slowed the engine to the point it would be tactically weak. But the ideas were known 10 years ago already, we just couldn't do them (actually, in Cray Blitz we did a lot of them as the hardware allowed us to get away with things that a non-vector machine would not.) >If by below you think software only counts for less then 30 elo a >year. In 10 years that only comes to 0 to 299 elo over 10 years. I think we can >show over 30 elo a year on the software side. If I played Fritz 2 a ten year >old program I know already the current programs will best Fritz by well over 300 >elo points. The results are just ugly we you play old vs new, but when you play >the very old programs well.... Play the old program vs the new program using old hardware. I'll bet you won't see 300 elo difference. I wouldn't be surprised if some of the older programs actually would come out on top, as they were so optimized to a specific processor speed. And notice that when I say "at least more than 50% of the strength increases come from hardware" that doesn't mean that just faster hardware is all that is needed. Sometimes the faster hardware makes it possible for us to do something in software that we could not afford on slower machines. Without the faster hardware the software feature would not be possible. And I still attribute that gain to hardware since without it it would be impossible to do. > > >> >>I believe that if you claim that each year the programs are 60 Elo stronger, >>that more than 30 of that Elo comes from hardware improvements. However, I >>have said, quite clearly, that the engines _have_ improved in steady increments. >> >>How that becomes "little software progress" is something my dictionary won't >>reveal. >> >> >> >> >>> >>>Well we can test this theory, Since we can run the old programs on modern >>>hardware and play them against the best programs of today. Then we will see just >>>how much or less hardware has to do with micro computer chess advancement. >>> >>> >> >> >>No. How about new programs on _old_ hardware? That is a fairer test. Old >>programs made compromises that were necessary because of the speed of the >>hardware they had. Those programs on faster hardware won't be as well-tuned >>for the faster hardware as new programs designed for such speeds. >> >>And don't just play comp vs comp. You need humans. Comp vs comp exaggerates >>the Elo difference between two engines. Playing on ICC would be a better >>test... >> >>> >>> >>> >>>> >>>>Mogens.
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.