Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: 2887-2970 kns in september

Author: Gordon Rattray

Date: 08:54:49 03/23/01

Go up one level in this thread


On March 23, 2001 at 07:06:11, Slater Wold wrote:

[snip]

>I too am an avid (street) racer, so I know where you're coming from.  However, I
>have come to understand that in computer chess, horsepower does not equal
>hardware.  In other words, you can justify that 10HP = .1 of a second in the 1/4
>mile.  However, I believe the rule in computer chess is x2 mhz = 50 elo or so.
>If a program is 2500 on an P4 1.5Ghz and you dual it to x 2 1.5Ghz, you're only
>getting 50 elo.  Not a huge increase.  Well, not like going from 300HP to 600HP
>anyway.
>
>Ed, Robert, and all the programmers will tell you, there have been little
>advances in computer chess programming in the last 10 years. The difference
>between Fritz on a 486 DX-2 66 vs a P4 1.5Ghz would be one I think anyone would
>be able to see.
>
>Just my 2 cents.
>
>Slate


I agree.

As a reader of CCC and a computer chess enthusiast, I constantly switch my
opinion of what it means to have the latest hardware for running chess programs.

Sometimes I get carried away with nodes per second, Ghz values, or hash table
sizes.  For some crazy reason it delights me to see, for example, Fritz
averaging over one million nodes per second...  ;-)

Thankfully however I also have "sanity spells" where I look beyond the bare
figures and think about what it all means in terms of chess playing strength.  I
remind myself that:

- despite having some pretty clever algorithms to reduce the branching factor
(down to 5 or so?!), chess still involves an exponential search that determines
the time involved in searching deeper
- even if a program does search deeper, the benefit of seeing deeper gets
smaller for each ply deeper (as pointed out already)
- hash table size shows diminshing returns as the size goes up
- chess knowledge plays an essential part of any chess program

So, despite the seemingly rapid advances in computer hardware, the problem
presented by chess accelerates even quicker.

In previous messages, people have mentioned the human factor involved in this
issue.  This has included trying to justify to ourselves why we spend so much
money on the latest hardware and software releases even if the gains are
relatively small.  I'm guilty too!

My excuse is that it takes me too long to give each of my 6 chess programs a
decent time to analyse a test position, using one PC.  Therefore I needed
another one to get better throughput and why not aim for the latest hardware so
that it becomes obscolete in two months instead of only one month. ;-)  And now
that I've got two PCs, I can also play more engine matches, and so I'd like to
experiment with more different programs.  My program collection is growing and
the time to compare analysis for a test postion is beginning to take too long
once again, so...  ;-)  This helps keep my concious happy even if my bank
balance isn't.

Anyway, anyone know how much the dual Athlons are going to cost? ;-)

Gordon



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.