Author: Christophe Theron
Date: 17:42:26 02/05/00
Go up one level in this thread
On February 05, 2000 at 19:02:11, CLiebert wrote:
>On February 05, 2000 at 13:05:47, Christophe Theron wrote:
>
>>On February 05, 2000 at 06:21:59, CLiebert wrote:
>>
>>>>
>>>>>First of all let me say that my testing method is not for those of us who have
>>>>>several fast computers for their games. I have one 200MMX only.
>>>>>
>>>>>Both programs are put in equally bad, but constant, conditions.
>>>
>>>This isn´t true. For example playing hiarcs against fritz with 16mb ram/ht
>>>is not the same. fritz suffers much more using small tables than hiarcs. The
>>>same case in a smaller way with tiger and the faster junior which needs more Ht
>>>s. I assume that tiger is a litte bit in favour playing under these conditions
>>>
>>>Christian
>>
>>
>>Fair testing is about giving each program the same amount of RESOURCE.
>>
>>That means same CPU, and same amount of RAM, amongst other things.
>
>On principle you are right.
>
>>If a program performs less well than its opponent with the same amount of RAM,
>>that's a design flaw. Why would you reward this program by giving it more hash
>>tables?
>>
>>So Didzis is right: the programs are in equally bad conditions.
>
>Ok. But I think these conditions are not suitable today. 60m/Games with
>16 MB RAM ...
It does not matter. You would need to have drastically different conditions to
notice any change in relative strength. Something like game in three days and
1Go hash tables.
If such a change ever happens.
>>Anyway, double the amount of hash tables for both programs, and all you'll see
>>is that their relative playing strength remains constant.
>
>Are you sure about it? That´s the main question: does an in/decrease of ht s
>has the same effect to every program (I would assume: no) !
Due to how hash tables are used by everybody, commercial programmers have found
how to use them very well, and I don't think anybody uses it much better than
his peers.
You can make the experiment if you want. On a large number of positions,
doubling the hash table size would give the same relative speedup for any top
commercial program.
I think this has been done before and the speedup for doubling the HT size is
6-7%, for any program.
I don't think a "full" hash table is an issue. A hash table is never "full". It
keeps on being useful as entries only get replaced.
If there is any difference because of this, the expected effect is probably a
2-3% speed loss, something you'll never be able to measure unless you play a
1000 games match.
Christophe
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.