Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Opinion on PC configuration?

Author: Tom Kerrigan

Date: 14:35:57 06/07/00

Go up one level in this thread


On June 07, 2000 at 15:44:14, Ricardo Gibert wrote:

>On June 07, 2000 at 15:07:58, Tom Kerrigan wrote:
>
>>On June 07, 2000 at 15:04:17, Eugene Nalimov wrote:
>>
>>>On June 07, 2000 at 13:24:19, Tom Kerrigan wrote:
>>>
>>>>On June 07, 2000 at 08:42:20, Laurence Chen wrote:
>>>>
>>>>>On June 07, 2000 at 05:16:33, Ricardo Gibert wrote:
>>>>>
>>>>>>On June 07, 2000 at 01:36:16, O. Veli wrote:
>>>>>>
>>>>>>>  I am planning on building a powerful yet cheap PC for chess. AFAIK there is no
>>>>>>>difference between Pentium III and Celeron chips on chess performance. A dual
>>>>>>>processor version is better than a single one so a dual Celeron + Deep Junior (
>>>>>>>and of course Crafty) will have a strong Elo/$ value (Dual or quad Pentium III
>>>>>>>is out of my reach). How much Elo would dual Celeron + Deep Junior gain compared
>>>>>>>to single Celeron + Junior? What other things should I keep in mind on this
>>>>>>>machine? Thanks.
>>>>>>
>>>>>>Your choice is not so simple. Consider:
>>>>>>
>>>>>>ABIT BP6 + 2 533 Celerons $350
>>>>>>
>>>>>>ASUS K7V + 1 700 Athlon   $350
>>>>>>
>>>>>>Its estimated that a doubling of speed is worth about a 50 rating point gain, so
>>>>>>we can use the following equation:
>>>>>>
>>>>>>73*ln(s1/s2) = Delta R
>>>>>>
>>>>>>73*ln(1066/700) = 31 rating points
>>>>>>
>>>>>>Now look at the SSDF ratings here:
>>>>>>
>>>>>>http://home3.swipnet.se/~w-36794/ssdf/nr000.htm
>>>>>>
>>>>>>Fritz 6  is rated 2721
>>>>>>
>>>>>>Junior 6 is rated 2689 (I'm assuming Deep Junior is the same on 1 cpu)
>>>>>>
>>>>>>2721 - 2689 = 32 rating points
>>>>>>
>>>>>>So someone with the Athlon will do just as well using Fritz 6 as you will using
>>>>>>the dual Celerons using Deep Junior. Of course, these are _very_ rough
>>>>>>calculations.
>>>>>>
>>>>>>We have assumed that the _relative_ ratings are accurate, that the 50 rating
>>>>>>point estimate is accurate, that Deep Junior is just as strong as Junior 6 on
>>>>>>one cpu, that Deep Junior scales up going to 2 cpus with 100% efficiency, etc.
>>>>>>Clearly, many of these assumptions aren't very reasonable.
>>>>>>
>>>>>>The dual cpu system is not everything it is cracked up to be as far as chess
>>>>>>goes. The quality of the program is just as important as the speed of the
>>>>>>hardware. But even if Deep Junior is just as good as Fritz 6, you do not gain
>>>>>>much.
>>>>>>
>>>>>>When you get a dual system, your choice of hardware is more limited and your
>>>>>>choice of software for it that takes advantage of it is more limited. You make
>>>>>>compromises. I prefer the ASUS motherboard to the ABIT motherboard. The ASUS
>>>>>>motherboard is the best one for the Athlon. The ABIT motherboard is not the best
>>>>>>for the Celeron. It's the one you get when you want to run a dual system.
>>>>>>
>>>>>>Naturally, there are other reasons for getting a dual system. I hope this helps
>>>>>>you make your decision.
>>>>>You forgot to mention about the OS, a dual processor will require Windows NT, or
>>>>>Windows 2000, since Microsoft no longer sells Windows NT 4, then you have no
>>>>>choice but buy Windows 2000, the OEM is about U.S. $350.00.
>>>>>IMHO I don't think that Celerons are suitable as Dual Processors for high end
>>>>>processing.  Celerons are great for Games, or some office suite programs such as
>>>>>Word Processing, and spreadsheets.
>>>>>Laurence
>>>>
>>>>Celerons are all but identical to Pentiums. I don't know why either chip would
>>>>be better at something, unless you're talking about some esoteric cache issue.
>>>>
>>>>-Tom
>>>
>>>Of course Celeron differs:
>>>
>>>(1) Smaller L2 cache size -- 128Kb vs. 256Kb (for newer PIIIs) or 512Kb at 1/2
>>>CPU speed (for PIIs and older PIIIs).
>>>(2) Slower bus speed -- 66MHz for older Celerons (vs. 100MHz for PII/PIII at
>>>that time), 100Mhz for newer ones (vs. 133MHz for new PIIIs).
>>>
>>>Both those factors hurts programs with high memory traffic in general, and
>>>multiprocessor systems in particular, as in "value computers segment" memory
>>>bandwith must be shared between CPUs. So, I'd not recommend to run CPU-bound
>>>database application on a dual Celeron system. But for chess program dual
>>>Celeron must be fine.
>>>
>>>Also notice that on dual-CPU system you often have better response time than on
>>>the single-CPU one (even when that single CPU is faster), so system can appear
>>>faster that it really is :-)
>>>
>>>Eugene
>>
>>Yes, there are some differences, but you'll only notice them if you pick your
>>apps carefully. For computer chess and everyday stuff, Celerons are the same as
>>Pentiums.
>>
>>And BTW, there's usually nothing wrong with running a Celeron at 100MHz FSB. :)
>>
>>-Tom
>
>I've recommended Celeron to a lot of people. It's a good cost effective CPU. My
>purpose in my post was to point out that going to a dual system will not create
>a "killer machine".
>
>People make to much out of NPS, when that is not nearly as important as EBF, a
>decent eval or good move ordering. Too many programmers here seem to try to tune
>the wrong things. Making a program run 10% faster will give 7 rating points. Big
>deal. It's hard to get that 10% and there isn't much of a payoff. Lowering the
>EBF significantly _is_ a big deal. Quality of the algorithm is more important
>than speed of the hardware. This is a property of combinatorially bounded
>algorithms.

At the same time, it is necessary to improve all aspects of a chess program
simultaneously if you want to get anywhere.

I have seen a number of people say, "I'm going to finish my eval function, then
I can work on the rest of my program." They get nowhere.

So, if a person is trying to improve their branching factor, and they can't
really think of a way to do it, they should work for the speed optimizations
instead of just sitting and scratching their head.

-Tom



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.