Author: Sune Fischer
Date: 03:08:21 04/15/02
Go up one level in this thread
On April 15, 2002 at 02:14:42, David Dory wrote: >On April 14, 2002 at 10:08:18, Sune Fischer wrote: > >>Besides, there has been software development into how to maximize the use of >the hash, so I'd say a good deal of the improvement here is actually better >software design. > >What did they improve the hash performance by? 2 whole percent? Woooowee!! > >It's a paltry amount. NOTHING like the improvements made in computer HARDWARE. In theory bigger is better, in practise you don't get much improvement beyond a certain size. But how do you know its only 2%, do you have any idea how Christophe got Tiger 200 elo stronger than Crafty? They are both using the same alpha-beta, so how could one possibly be 200 elo stronger, explain that! The countless number of hours put into testing and analysing, you don't think that is going to pay off? >Talk to Bob, or Christophe or Ed, J. Schaeffer, just anybody, Sune. Don't take >my word for it. Ken Thompson once described watching his program competing, and >just wanting to CRY for the lack of speed in the hardware. That's why he built >Belle, to overcome the HARDWARE limitations of PC's in the 70's. Hardware _is_ a factor, but not the only factor, which you seem to think. >Crayblitz didn't run on a Cray because they liked that snappy name. CHESS wasn't >a dominant champ because it ran on a PC. Fritz won't be fighting Kramnik from a >slow PC. That's because they know they'd get their butts kicked by Kramnik if >they came up to THE BATTLE with a run of the mill type of PC. > >Name one algorithim used today, that wasn't used 20 years ago? You don't have to reinvent the wheel every day to build a faster car....! You will have to ask the top programmers for examples, they are the ones to break new ground every day. >Null move is one noteable enhancement to A-B, but that's about it. Everything >else has been more knowledge and speed, ALL MADE POSSIBLE (and practical), >because of THE COMPUTER HARDWARE IMPROVEMENTS. > >Re: The need for big, fast, hard disks > >You can't keep a lot of EGTB's in memory, Sune. You need big, fast, hard disks, >and all the main memory you can get. If you get enough memory, you can page the >part of the EGTB's into memory, so you don't have to keep going out for a disk >read all the time (which takes 5-10 THOUSAND times longer)!. NOW THERE'S A BIG >IMPROVEMENT!! Ha! First of all you need to reach the EGTBs, no improvement in opening or middelgame. I remember Bob told me once, that engines search very deep in the endgame, because of the few pieces the transposition table is very effective, so the EGTBs is no factor 15000 (whatever you mean by that BTW). >BTW, EGTB's are NOT a software improvement of the past 20 years, (Ken Thompson >had his in 1977,) but they're a BIG help! >PC programs couldn't do this 20 years ago because they didn't have the speed or >the memory for it. Another hardware bottleneck we've overcome! And everytime hardware improves, some software has to be invented and finetuned to take advantage of it. When we get quantum computers I'm sure the alpha-beta will recieve quite a make-over. I believe the software has to grow along with the hardware, if it was "just hardware" what have the programmers been doing for the last 20 years? >>Think again please. I'd say it is _at least_ 50% software progress now >>a days. > >Done. I'd say you don't know what the - well, I can be a gentleman and say I >think you're more attuned to the small increases in software, and have not >particularly noticed the HUGE increases in hardware performance. > >David So I take it you still believe programs are 15,000 times stronger because of faster harddrives? Well, at least I tried... ;) -S.
This page took 0.01 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.